Added conversation history

Replace Ollama embedding function with OpenAI's text-embedding-3-small
model for improved embedding quality and consistency.

🤖 Generated with [Claude Code](https://claude.com/claude-code)

Co-Authored-By: Claude <noreply@anthropic.com>

.dockerignore

@@ -0,0 +1,16 @@
+.git
+.gitignore
+README.md
+.env
+.DS_Store
+chromadb/
+chroma_db/
+raggr-frontend/node_modules/
+__pycache__/
+*.pyc
+*.pyo
+*.pyd
+.Python
+.venv/
+venv/
+.pytest_cache/

.python-version

@@ -0,0 +1 @@
+3.13

Dockerfile

@@ -0,0 +1,46 @@
+FROM python:3.13-slim
+
+WORKDIR /app
+
+# Install system dependencies, Node.js, Yarn, and uv
+RUN apt-get update && apt-get install -y \
+    build-essential \
+    curl \
+    && curl -fsSL https://deb.nodesource.com/setup_20.x | bash - \
+    && apt-get install -y nodejs \
+    && npm install -g yarn \
+    && rm -rf /var/lib/apt/lists/* \
+    && curl -LsSf https://astral.sh/uv/install.sh | sh
+
+# Add uv to PATH
+ENV PATH="/root/.local/bin:$PATH"
+
+# Copy dependency files
+COPY pyproject.toml ./
+
+# Install Python dependencies using uv
+RUN uv pip install --system -e .
+
+# Copy application code
+COPY *.py ./
+COPY startup.sh ./
+RUN chmod +x startup.sh
+
+# Copy frontend code and build
+COPY raggr-frontend ./raggr-frontend
+WORKDIR /app/raggr-frontend
+RUN yarn install && yarn build
+WORKDIR /app
+
+# Create ChromaDB directory
+RUN mkdir -p /app/chromadb
+
+# Expose port
+EXPOSE 8080
+
+# Set environment variables
+ENV PYTHONPATH=/app
+ENV CHROMADB_PATH=/app/chromadb
+
+# Run the startup script
+CMD ["./startup.sh"]

app.py

@@ -0,0 +1,102 @@
+import os
+
+from quart import Quart, request, jsonify, render_template, send_from_directory
+from tortoise.contrib.quart import register_tortoise
+
+from quart_jwt_extended import JWTManager
+
+from main import consult_simba_oracle
+from blueprints.conversation.logic import (
+    get_the_only_conversation,
+    add_message_to_conversation,
+)
+
+app = Quart(
+    __name__,
+    static_folder="raggr-frontend/dist/static",
+    template_folder="raggr-frontend/dist",
+)
+
+app.config["JWT_SECRET_KEY"] = os.getenv("JWT_SECRET_KEY", "SECRET_KEY")
+jwt = JWTManager(app)
+
+# Initialize Tortoise ORM
+register_tortoise(
+    app,
+    db_url=os.getenv("DATABASE_URL", "sqlite://raggr.db"),
+    modules={"models": ["blueprints.conversation.models"]},
+    generate_schemas=True,
+)
+
+
+# Serve React static files
+@app.route("/static/<path:filename>")
+async def static_files(filename):
+    return await send_from_directory(app.static_folder, filename)
+
+
+# Serve the React app for all routes (catch-all)
+@app.route("/", defaults={"path": ""})
+@app.route("/<path:path>")
+async def serve_react_app(path):
+    if path and os.path.exists(os.path.join(app.template_folder, path)):
+        return await send_from_directory(app.template_folder, path)
+    return await render_template("index.html")
+
+
+@app.route("/api/query", methods=["POST"])
+async def query():
+    data = await request.get_json()
+    query = data.get("query")
+    # add message to database
+    conversation = await get_the_only_conversation()
+    print(conversation)
+    await add_message_to_conversation(
+        conversation=conversation, message=query, speaker="user"
+    )
+
+    response = consult_simba_oracle(query)
+    await add_message_to_conversation(
+        conversation=conversation, message=response, speaker="simba"
+    )
+    return jsonify({"response": response})
+
+
+@app.route("/api/messages", methods=["GET"])
+async def get_messages():
+    conversation = await get_the_only_conversation()
+    # Prefetch related messages
+    await conversation.fetch_related("messages")
+
+    # Manually serialize the conversation with messages
+    messages = []
+    for msg in conversation.messages:
+        messages.append(
+            {
+                "id": str(msg.id),
+                "text": msg.text,
+                "speaker": msg.speaker.value,
+                "created_at": msg.created_at.isoformat(),
+            }
+        )
+
+    return jsonify(
+        {
+            "id": str(conversation.id),
+            "name": conversation.name,
+            "messages": messages,
+            "created_at": conversation.created_at.isoformat(),
+            "updated_at": conversation.updated_at.isoformat(),
+        }
+    )
+
+
+# @app.route("/api/ingest", methods=["POST"])
+# def webhook():
+# data = request.get_json()
+# print(data)
+# return jsonify({"status": "received"})
+
+
+if __name__ == "__main__":
+    app.run(host="0.0.0.0", port=8080, debug=True)

blueprints/conversation/__init__.py

@@ -0,0 +1,17 @@
+from quart import Blueprint, jsonify
+from .models import (
+    Conversation,
+    PydConversation,
+)
+
+conversation_blueprint = Blueprint(
+    "conversation_api", __name__, url_prefix="/api/conversation"
+)
+
+
+@conversation_blueprint.route("/<conversation_id>")
+async def get_conversation(conversation_id: str):
+    conversation = await Conversation.get(id=conversation_id)
+    serialized_conversation = await PydConversation.from_tortoise_orm(conversation)
+
+    return jsonify(serialized_conversation.model_dump_json())

blueprints/conversation/logic.py

@@ -0,0 +1,32 @@
+from .models import Conversation, ConversationMessage
+
+
+async def create_conversation(name: str = "") -> Conversation:
+    conversation = await Conversation.create(name=name)
+    return conversation
+
+
+async def add_message_to_conversation(
+    conversation: Conversation,
+    message: str,
+    speaker: str,
+) -> ConversationMessage:
+    print(conversation, message, speaker)
+    message = await ConversationMessage.create(
+        text=message,
+        speaker=speaker,
+        conversation=conversation,
+    )
+
+    return message
+
+
+async def get_the_only_conversation() -> Conversation:
+    try:
+        conversation = await Conversation.all().first()
+        if conversation is None:
+            conversation = await Conversation.create(name="simba_chat")
+    except Exception as _e:
+        conversation = await Conversation.create(name="simba_chat")
+
+    return conversation

blueprints/conversation/models.py

@@ -0,0 +1,41 @@
+import enum
+
+from tortoise.models import Model
+from tortoise import fields
+from tortoise.contrib.pydantic import (
+    pydantic_queryset_creator,
+    pydantic_model_creator,
+)
+
+
+class Speaker(enum.Enum):
+    USER = "user"
+    SIMBA = "simba"
+
+
+class Conversation(Model):
+    id = fields.UUIDField(primary_key=True)
+    name = fields.CharField(max_length=255)
+    created_at = fields.DatetimeField(auto_now_add=True)
+    updated_at = fields.DatetimeField(auto_now=True)
+
+    class Meta:
+        table = "conversations"
+
+
+class ConversationMessage(Model):
+    id = fields.UUIDField(primary_key=True)
+    text = fields.TextField()
+    conversation = fields.ForeignKeyField(
+        "models.Conversation", related_name="messages"
+    )
+    created_at = fields.DatetimeField(auto_now_add=True)
+    speaker = fields.CharEnumField(enum_type=Speaker, max_length=10)
+
+    class Meta:
+        table = "conversation_messages"
+
+
+PydConversationMessage = pydantic_model_creator(ConversationMessage)
+PydConversation = pydantic_model_creator(Conversation, name="Conversation")
+PydListConversationMessage = pydantic_queryset_creator(ConversationMessage)

chunker.py

@@ -0,0 +1,142 @@
+import os
+from math import ceil
+import re
+from typing import Union
+from uuid import UUID, uuid4
+from ollama import Client
+from chromadb.utils.embedding_functions.openai_embedding_function import (
+    OpenAIEmbeddingFunction,
+)
+from dotenv import load_dotenv
+from llm import LLMClient
+
+
+load_dotenv()
+
+ollama_client = Client(
+    host=os.getenv("OLLAMA_HOST", "http://localhost:11434"), timeout=10.0
+)
+
+
+def remove_headers_footers(text, header_patterns=None, footer_patterns=None):
+    if header_patterns is None:
+        header_patterns = [r"^.*Header.*$"]
+    if footer_patterns is None:
+        footer_patterns = [r"^.*Footer.*$"]
+
+    for pattern in header_patterns + footer_patterns:
+        text = re.sub(pattern, "", text, flags=re.MULTILINE)
+
+    return text.strip()
+
+
+def remove_special_characters(text, special_chars=None):
+    if special_chars is None:
+        special_chars = r"[^A-Za-z0-9\s\.,;:\'\"\?\!\-]"
+
+    text = re.sub(special_chars, "", text)
+    return text.strip()
+
+
+def remove_repeated_substrings(text, pattern=r"\.{2,}"):
+    text = re.sub(pattern, ".", text)
+    return text.strip()
+
+
+def remove_extra_spaces(text):
+    text = re.sub(r"\n\s*\n", "\n\n", text)
+    text = re.sub(r"\s+", " ", text)
+
+    return text.strip()
+
+
+def preprocess_text(text):
+    # Remove headers and footers
+    text = remove_headers_footers(text)
+
+    # Remove special characters
+    text = remove_special_characters(text)
+
+    # Remove repeated substrings like dots
+    text = remove_repeated_substrings(text)
+
+    # Remove extra spaces between lines and within lines
+    text = remove_extra_spaces(text)
+
+    # Additional cleaning steps can be added here
+
+    return text.strip()
+
+
+class Chunk:
+    def __init__(
+        self,
+        text: str,
+        size: int,
+        document_id: UUID,
+        chunk_id: int,
+        embedding,
+    ):
+        self.text = text
+        self.size = size
+        self.document_id = document_id
+        self.chunk_id = chunk_id
+        self.embedding = embedding
+
+
+class Chunker:
+    def __init__(self, collection) -> None:
+        self.collection = collection
+        self.llm_client = LLMClient()
+
+    def embedding_fx(self, inputs):
+        openai_embedding_fx = OpenAIEmbeddingFunction(
+            api_key=os.getenv("OPENAI_API_KEY"),
+            model_name="text-embedding-3-small",
+        )
+        return openai_embedding_fx(inputs)
+
+    def chunk_document(
+        self,
+        document: str,
+        chunk_size: int = 1000,
+        metadata: dict[str, Union[str, float]] = {},
+    ) -> list[Chunk]:
+        doc_uuid = uuid4()
+
+        chunk_size = min(chunk_size, len(document)) or 1
+
+        chunks = []
+        num_chunks = ceil(len(document) / chunk_size)
+        document_length = len(document)
+
+        for i in range(num_chunks):
+            curr_pos = i * num_chunks
+            to_pos = (
+                curr_pos + chunk_size
+                if curr_pos + chunk_size < document_length
+                else document_length
+            )
+            text_chunk = self.clean_document(document[curr_pos:to_pos])
+
+            embedding = self.embedding_fx([text_chunk])
+            self.collection.add(
+                ids=[str(doc_uuid) + ":" + str(i)],
+                documents=[text_chunk],
+                embeddings=embedding,
+                metadatas=[metadata],
+            )
+
+        return chunks
+
+    def clean_document(self, document: str) -> str:
+        """This function will remove information that is noise or already known.
+
+        Example: We already know all the things in here are Simba-related, so we don't need things like
+        "Sumamry of simba's visit"
+        """
+
+        document = document.replace("\\n", "")
+        document = document.strip()
+
+        return preprocess_text(document)

cleaner.py

@@ -0,0 +1,165 @@
+import os
+import sys
+import tempfile
+
+import argparse
+from dotenv import load_dotenv
+import ollama
+from PIL import Image
+import fitz
+
+from request import PaperlessNGXService
+
+load_dotenv()
+
+# Configure ollama client with URL from environment or default to localhost
+ollama_client = ollama.Client(host=os.getenv("OLLAMA_URL", "http://localhost:11434"))
+
+parser = argparse.ArgumentParser(description="use llm to clean documents")
+parser.add_argument("document_id", type=str, help="questions about simba's health")
+
+
+def pdf_to_image(filepath: str, dpi=300) -> list[str]:
+    """Returns the filepaths to the created images"""
+    image_temp_files = []
+    try:
+        pdf_document = fitz.open(filepath)
+        print(f"\nConverting '{os.path.basename(filepath)}' to temporary images...")
+
+        for page_num in range(len(pdf_document)):
+            page = pdf_document.load_page(page_num)
+            zoom = dpi / 72
+            mat = fitz.Matrix(zoom, zoom)
+            pix = page.get_pixmap(matrix=mat)
+
+            # Create a temporary file for the image. delete=False is crucial.
+            with tempfile.NamedTemporaryFile(
+                delete=False,
+                suffix=".png",
+                prefix=f"pdf_page_{page_num + 1}_",
+            ) as temp_image_file:
+                temp_image_path = temp_image_file.name
+
+            # Save the pixel data to the temporary file
+            pix.save(temp_image_path)
+            image_temp_files.append(temp_image_path)
+            print(
+                f"  -> Saved page {page_num + 1} to temporary file: '{temp_image_path}'"
+            )
+
+        print("\nConversion successful! ✨")
+        return image_temp_files
+
+    except Exception as e:
+        print(f"An error occurred during PDF conversion: {e}", file=sys.stderr)
+        # Clean up any image files that were created before the error
+        for path in image_temp_files:
+            os.remove(path)
+        return []
+
+
+def merge_images_vertically_to_tempfile(image_paths):
+    """
+    Merges a list of images vertically and saves the result to a temporary file.
+
+    Args:
+        image_paths (list): A list of strings, where each string is the
+                            filepath to an image.
+
+    Returns:
+        str: The filepath of the temporary merged image file.
+    """
+    if not image_paths:
+        print("Error: The list of image paths is empty.")
+        return None
+
+    # Open all images and check for consistency
+    try:
+        images = [Image.open(path) for path in image_paths]
+    except FileNotFoundError as e:
+        print(f"Error: Could not find image file: {e}")
+        return None
+
+    widths, heights = zip(*(img.size for img in images))
+    max_width = max(widths)
+
+    # All images must have the same width
+    if not all(width == max_width for width in widths):
+        print("Warning: Images have different widths. They will be resized.")
+        resized_images = []
+        for img in images:
+            if img.size[0] != max_width:
+                img = img.resize(
+                    (max_width, int(img.size[1] * (max_width / img.size[0])))
+                )
+            resized_images.append(img)
+        images = resized_images
+        heights = [img.size[1] for img in images]
+
+    # Calculate the total height of the merged image
+    total_height = sum(heights)
+
+    # Create a new blank image with the combined dimensions
+    merged_image = Image.new("RGB", (max_width, total_height))
+
+    # Paste each image onto the new blank image
+    y_offset = 0
+    for img in images:
+        merged_image.paste(img, (0, y_offset))
+        y_offset += img.height
+
+    # Create a temporary file and save the image
+    temp_file = tempfile.NamedTemporaryFile(suffix=".png", delete=False)
+    temp_path = temp_file.name
+    merged_image.save(temp_path)
+    temp_file.close()
+
+    print(f"Successfully merged {len(images)} images into temporary file: {temp_path}")
+    return temp_path
+
+
+OCR_PROMPT = """
+    You job is to extract text from the images I provide you. Extract every bit of the text in the image. Don't say anything just do your job. Text should be same as in the images. If there are multiple images, categorize the transcriptions by page.
+
+Things to avoid:
+- Don't miss anything to extract from the images
+
+Things to include:
+- Include everything, even anything inside [], (), {} or anything.
+- Include any repetitive things like "..." or anything
+- If you think there is any mistake in image just include it too
+
+Someone will kill the innocent kittens if you don't extract the text exactly. So, make sure you extract every bit of the text. Only output the extracted text.
+"""
+
+
+def summarize_pdf_image(filepaths: list[str]):
+    res = ollama_client.chat(
+        model="gemma3:4b",
+        messages=[
+            {
+                "role": "user",
+                "content": OCR_PROMPT,
+                "images": filepaths,
+            }
+        ],
+    )
+
+    return res["message"]["content"]
+
+
+if __name__ == "__main__":
+    args = parser.parse_args()
+    ppngx = PaperlessNGXService()
+
+    if args.document_id:
+        doc_id = args.document_id
+        file = ppngx.get_doc_by_id(doc_id=doc_id)
+        pdf_path = ppngx.download_pdf_from_id(doc_id)
+        print(pdf_path)
+        image_paths = pdf_to_image(filepath=pdf_path)
+        summary = summarize_pdf_image(filepaths=image_paths)
+        print(summary)
+        file["content"] = summary
+        print(file)
+        ppngx.upload_cleaned_content(doc_id, file)

docker-compose.yml

@@ -0,0 +1,17 @@
+version: "3.8"
+
+services:
+  raggr:
+    image: torrtle/simbarag:latest
+    network_mode: host
+    environment:
+      - PAPERLESS_TOKEN=${PAPERLESS_TOKEN}
+      - BASE_URL=${BASE_URL}
+      - OLLAMA_URL=${OLLAMA_URL:-http://localhost:11434}
+      - CHROMADB_PATH=/app/chromadb
+      - OPENAI_API_KEY=${OPENAI_API_KEY}
+    volumes:
+      - chromadb_data:/app/chromadb
+
+volumes:
+  chromadb_data:

image_process.py

@@ -0,0 +1,83 @@
+from ollama import Client
+import argparse
+import os
+import logging
+from PIL import Image, ExifTags
+from pillow_heif import register_heif_opener
+from pydantic import BaseModel
+
+from dotenv import load_dotenv
+
+load_dotenv()
+
+register_heif_opener()
+
+logging.basicConfig(level=logging.INFO)
+
+
+parser = argparse.ArgumentParser(
+    prog="SimbaImageProcessor",
+    description="What the program does",
+    epilog="Text at the bottom of help",
+)
+
+parser.add_argument("filepath")
+
+client = Client(host=os.getenv("OLLAMA_HOST", "http://localhost:11434"))
+
+
+class SimbaImageDescription(BaseModel):
+    image_date: str
+    description: str
+
+
+def describe_simba_image(input):
+    logging.info("Opening image of Simba ...")
+    if "heic" in input.lower() or "heif" in input.lower():
+        new_filepath = input.split(".")[0] + ".jpg"
+        img = Image.open(input)
+        img.save(new_filepath, "JPEG")
+        logging.info("Extracting EXIF...")
+        exif = {
+            ExifTags.TAGS[k]: v for k, v in img.getexif().items() if k in ExifTags.TAGS
+        }
+        img = Image.open(new_filepath)
+        input = new_filepath
+    else:
+        img = Image.open(input)
+
+        logging.info("Extracting EXIF...")
+        exif = {
+            ExifTags.TAGS[k]: v for k, v in img.getexif().items() if k in ExifTags.TAGS
+        }
+
+    if "MakerNote" in exif:
+        exif.pop("MakerNote")
+
+    logging.info(exif)
+
+    prompt = f"Simba is an orange cat belonging to Ryan Chen. In 2025, they lived in New York. In 2024, they lived in California. Analyze the following image and tell me what Simba seems to be doing. Be extremely descriptive about Simba, things in the background, and the setting of the image. I will also include the EXIF data of the image, please use it to help you determine information about Simba. EXIF: {exif}. Put the notes in the description field and the date in the image_date field."
+
+    logging.info("Sending info to Ollama ...")
+    response = client.chat(
+        model="gemma3:4b",
+        messages=[
+            {
+                "role": "system",
+                "content": "you are a very shrewd and descriptive note taker. all of your responses will be formatted like notes in bullet points. be very descriptive. do not leave a single thing out.",
+            },
+            {"role": "user", "content": prompt, "images": [input]},
+        ],
+        format=SimbaImageDescription.model_json_schema(),
+    )
+
+    result = SimbaImageDescription.model_validate_json(response["message"]["content"])
+
+    return result
+
+
+if __name__ == "__main__":
+    args = parser.parse_args()
+    if args.filepath:
+        logging.info
+        describe_simba_image(input=args.filepath)

index_immich.py

@@ -0,0 +1,115 @@
+import httpx
+import os
+from pathlib import Path
+import logging
+import tempfile
+
+from image_process import describe_simba_image
+from request import PaperlessNGXService
+import sqlite3
+
+logging.basicConfig(level=logging.INFO)
+
+
+from dotenv import load_dotenv
+
+load_dotenv()
+
+# Configuration from environment variables
+IMMICH_URL = os.getenv("IMMICH_URL", "http://localhost:2283")
+API_KEY = os.getenv("IMMICH_API_KEY")
+PERSON_NAME = os.getenv("PERSON_NAME", "Simba")  # Name of the tagged person/pet
+DOWNLOAD_DIR = os.getenv("DOWNLOAD_DIR", "./simba_photos")
+
+# Set up headers
+headers = {"x-api-key": API_KEY, "Content-Type": "application/json"}
+
+VISITED = {}
+
+if __name__ == "__main__":
+    conn = sqlite3.connect("./visited.db")
+    c = conn.cursor()
+    c.execute("select immich_id from visited")
+    rows = c.fetchall()
+    for row in rows:
+        VISITED.add(row[0])
+
+    ppngx = PaperlessNGXService()
+    people_url = f"{IMMICH_URL}/api/search/person?name=Simba"
+    people = httpx.get(people_url, headers=headers).json()
+
+    simba_id = people[0]["id"]
+
+    ids = {}
+
+    asset_search = f"{IMMICH_URL}/api/search/smart"
+    request_body = {"query": "orange cat"}
+    results = httpx.post(asset_search, headers=headers, json=request_body)
+
+    assets = results.json()["assets"]
+    for asset in assets["items"]:
+        if asset["type"] == "IMAGE" and asset["id"] not in VISITED:
+            ids[asset["id"]] = asset.get("originalFileName")
+    nextPage = assets.get("nextPage")
+
+    # while nextPage != None:
+    # logging.info(f"next page: {nextPage}")
+    # request_body["page"] = nextPage
+    # results = httpx.post(asset_search, headers=headers, json=request_body)
+    # assets = results.json()["assets"]
+
+    # for asset in assets["items"]:
+    # if asset["type"] == "IMAGE":
+    # ids.add(asset['id'])
+
+    # nextPage = assets.get("nextPage")
+
+    asset_search = f"{IMMICH_URL}/api/search/smart"
+    request_body = {"query": "simba"}
+    results = httpx.post(asset_search, headers=headers, json=request_body)
+    for asset in results.json()["assets"]["items"]:
+        if asset["type"] == "IMAGE":
+            ids[asset["id"]] = asset.get("originalFileName")
+
+    for immich_asset_id, immich_filename in ids.items():
+        try:
+            response = httpx.get(
+                f"{IMMICH_URL}/api/assets/{immich_asset_id}/original", headers=headers
+            )
+
+            path = os.path.join("/Users/ryanchen/Programs/raggr", immich_filename)
+            file = open(path, "wb+")
+            for chunk in response.iter_bytes(chunk_size=8192):
+                file.write(chunk)
+
+            logging.info("Processing image ...")
+            description = describe_simba_image(path)
+
+            image_description = description.description
+            image_date = description.image_date
+
+            description_filepath = os.path.join(
+                "/Users/ryanchen/Programs/raggr", f"SIMBA_DESCRIBE_001.txt"
+            )
+            file = open(description_filepath, "w+")
+            file.write(image_description)
+            file.close()
+
+            file = open(description_filepath, "rb")
+            ppngx.upload_description(
+                description_filepath=description_filepath,
+                file=file,
+                title="SIMBA_DESCRIBE_001.txt",
+                exif_date=image_date,
+            )
+            file.close()
+
+            c.execute("INSERT INTO visited (immich_id) values (?)", (immich_asset_id,))
+            conn.commit()
+            logging.info("Processing complete. Deleting file.")
+            os.remove(file.name)
+        except Exception as e:
+            logging.info(f"something went wrong for {immich_filename}")
+            logging.info(e)
+
+    conn.close()

llm.py

@@ -0,0 +1,64 @@
+import os
+
+from ollama import Client
+from openai import OpenAI
+
+import logging
+
+logging.basicConfig(level=logging.INFO)
+
+
+class LLMClient:
+    def __init__(self):
+        try:
+            self.ollama_client = Client(
+                host=os.getenv("OLLAMA_URL", "http://localhost:11434"), timeout=10.0
+            )
+            self.ollama_client.chat(
+                model="gemma3:4b", messages=[{"role": "system", "content": "test"}]
+            )
+            self.PROVIDER = "ollama"
+            logging.info("Using Ollama as LLM backend")
+        except Exception as e:
+            print(e)
+            self.openai_client = OpenAI()
+            self.PROVIDER = "openai"
+            logging.info("Using OpenAI as LLM backend")
+
+    def chat(
+        self,
+        prompt: str,
+        system_prompt: str,
+    ):
+        if self.PROVIDER == "ollama":
+            response = self.ollama_client.chat(
+                model="gemma3:4b",
+                messages=[
+                    {
+                        "role": "system",
+                        "content": system_prompt,
+                    },
+                    {"role": "user", "content": prompt},
+                ],
+            )
+            print(response)
+            output = response.message.content
+        elif self.PROVIDER == "openai":
+            response = self.openai_client.responses.create(
+                model="gpt-4o-mini",
+                input=[
+                    {
+                        "role": "system",
+                        "content": system_prompt,
+                    },
+                    {"role": "user", "content": prompt},
+                ],
+            )
+            output = response.output_text
+
+        return output
+
+
+if __name__ == "__main__":
+    client = Client()
+    client.chat(model="gemma3:4b", messages=[{"role": "system", "promp": "hack"}])

main.py

@@ -1,83 +1,243 @@
+import datetime
+import logging
+import os
+import sqlite3
+
+import argparse
+import chromadb
 import ollama
-from uuid import uuid4, UUID
+
 
 from request import PaperlessNGXService
-
-from math import ceil
-
-import chromadb
-
-client = chromadb.EphemeralClient()
-collection = client.create_collection(name="docs")
+from chunker import Chunker
+from cleaner import pdf_to_image, summarize_pdf_image
+from llm import LLMClient
+from query import QueryGenerator
 
 
-class Chunk:
-    def __init__(
-        self,
-        text: str,
-        size: int,
-        document_id: UUID,
-        chunk_id: int,
-        embedding,
-    ):
-        self.text = text
-        self.size = size
-        self.document_id = document_id
-        self.chunk_id = chunk_id
-        self.embedding = embedding
+from dotenv import load_dotenv
 
+_dotenv_loaded = load_dotenv()
 
-class Chunker:
-    def __init__(self) -> None:
-        pass
-
-    def chunk_document(self, document: str, chunk_size: int = 300) -> list[Chunk]:
-        doc_uuid = uuid4()
-
-        chunks = []
-        num_chunks = ceil(len(document) / chunk_size)
-        document_length = len(document)
-
-        for i in range(num_chunks):
-            curr_pos = i * num_chunks
-            to_pos = (
-                curr_pos + num_chunks
-                if curr_pos + num_chunks < document_length
-                else document_length
-            )
-            text_chunk = document[curr_pos:to_pos]
-
-            collection.add(
-                ids=[str(doc_uuid) + ":" + str(i)],
-                documents=[text_chunk],
-            )
-
-        return chunks
-
-
-# Setup
-
-# Step 1: Get the text
-ppngx = PaperlessNGXService()
-docs = ppngx.get_data()
-texts = [doc["content"] for doc in docs]
-
-# Step 2: Create chunks
-chunker = Chunker()
-
-print(f"chunking {len(texts)} documents")
-for text in texts:
-    chunker.chunk_document(document=text)
-
-# Ask
-input = "How many teeth has Simba had removed?"
-response = ollama.embed(model="mxbai-embed-large", input=input)
-results = collection.query(query_texts=[input], n_results=1)
-print(results)
-# Generate
-output = ollama.generate(
-    model="gemma3n:e4b",
-    prompt=f"Using this data: {results}. Respond to this prompt: {input}",
+# Configure ollama client with URL from environment or default to localhost
+ollama_client = ollama.Client(
+    host=os.getenv("OLLAMA_URL", "http://localhost:11434"), timeout=10.0
 )
 
-print(output["response"])
+client = chromadb.PersistentClient(path=os.getenv("CHROMADB_PATH", ""))
+simba_docs = client.get_or_create_collection(name="simba_docs2")
+feline_vet_lookup = client.get_or_create_collection(name="feline_vet_lookup")
+
+parser = argparse.ArgumentParser(
+    description="An LLM tool to query information about Simba <3"
+)
+
+parser.add_argument("query", type=str, help="questions about simba's health")
+parser.add_argument(
+    "--reindex", action="store_true", help="re-index the simba documents"
+)
+parser.add_argument("--index", help="index a file")
+
+ppngx = PaperlessNGXService()
+
+llm_client = LLMClient()
+
+
+def index_using_pdf_llm(doctypes):
+    logging.info("reindex data...")
+    files = ppngx.get_data()
+    for file in files:
+        document_id: int = file["id"]
+        pdf_path = ppngx.download_pdf_from_id(id=document_id)
+        image_paths = pdf_to_image(filepath=pdf_path)
+        logging.info(f"summarizing {file}")
+        generated_summary = summarize_pdf_image(filepaths=image_paths)
+        file["content"] = generated_summary
+
+    chunk_data(files, simba_docs, doctypes=doctypes)
+
+
+def date_to_epoch(date_str: str) -> float:
+    split_date = date_str.split("-")
+    date = datetime.datetime(
+        int(split_date[0]),
+        int(split_date[1]),
+        int(split_date[2]),
+        0,
+        0,
+        0,
+    )
+
+    return date.timestamp()
+
+
+def chunk_data(docs, collection, doctypes):
+    # Step 2: Create chunks
+    chunker = Chunker(collection)
+
+    logging.info(f"chunking {len(docs)} documents")
+    texts: list[str] = [doc["content"] for doc in docs]
+    with sqlite3.connect("visited.db") as conn:
+        to_insert = []
+        c = conn.cursor()
+        for index, text in enumerate(texts):
+            metadata = {
+                "created_date": date_to_epoch(docs[index]["created_date"]),
+                "filename": docs[index]["original_file_name"],
+                "document_type": doctypes.get(docs[index]["document_type"], ""),
+            }
+
+            if doctypes:
+                metadata["type"] = doctypes.get(docs[index]["document_type"])
+
+            chunker.chunk_document(
+                document=text,
+                metadata=metadata,
+            )
+            to_insert.append((docs[index]["id"],))
+
+        c.executemany(
+            "INSERT INTO indexed_documents (paperless_id) values (?)", to_insert
+        )
+        conn.commit()
+
+
+def chunk_text(texts: list[str], collection):
+    chunker = Chunker(collection)
+
+    for index, text in enumerate(texts):
+        metadata = {}
+        chunker.chunk_document(
+            document=text,
+            metadata=metadata,
+        )
+
+
+def consult_oracle(input: str, collection):
+    import time
+
+    chunker = Chunker(collection)
+
+    start_time = time.time()
+
+    # Ask
+    logging.info("Starting query generation")
+    qg_start = time.time()
+    qg = QueryGenerator()
+    doctype_query = qg.get_doctype_query(input=input)
+    # metadata_filter = qg.get_query(input)
+    metadata_filter = {**doctype_query}
+    logging.info(metadata_filter)
+    qg_end = time.time()
+    logging.info(f"Query generation took {qg_end - qg_start:.2f} seconds")
+
+    logging.info("Starting embedding generation")
+    embedding_start = time.time()
+    embeddings = chunker.embedding_fx(inputs=[input])
+    embedding_end = time.time()
+    logging.info(
+        f"Embedding generation took {embedding_end - embedding_start:.2f} seconds"
+    )
+
+    logging.info("Starting collection query")
+    query_start = time.time()
+    results = collection.query(
+        query_texts=[input],
+        query_embeddings=embeddings,
+        where=metadata_filter,
+    )
+    query_end = time.time()
+    logging.info(f"Collection query took {query_end - query_start:.2f} seconds")
+
+    # Generate
+    logging.info("Starting LLM generation")
+    llm_start = time.time()
+    system_prompt = "You are a helpful assistant that understands veterinary terms."
+    prompt = f"Using the following data, help answer the user's query by providing as many details as possible. Using this data: {results}. Respond to this prompt: {input}"
+    output = llm_client.chat(prompt=prompt, system_prompt=system_prompt)
+    llm_end = time.time()
+    logging.info(f"LLM generation took {llm_end - llm_start:.2f} seconds")
+
+    total_time = time.time() - start_time
+    logging.info(f"Total consult_oracle execution took {total_time:.2f} seconds")
+
+    return output
+
+
+def paperless_workflow(input):
+    # Step 1: Get the text
+    ppngx = PaperlessNGXService()
+    docs = ppngx.get_data()
+
+    chunk_data(docs, collection=simba_docs)
+    consult_oracle(input, simba_docs)
+
+
+def consult_simba_oracle(input: str):
+    return consult_oracle(
+        input=input,
+        collection=simba_docs,
+    )
+
+
+def filter_indexed_files(docs):
+    with sqlite3.connect("visited.db") as conn:
+        c = conn.cursor()
+        c.execute(
+            "CREATE TABLE IF NOT EXISTS indexed_documents (id INTEGER PRIMARY KEY AUTOINCREMENT, paperless_id INTEGER)"
+        )
+        c.execute("SELECT paperless_id FROM indexed_documents")
+        rows = c.fetchall()
+        conn.commit()
+
+    visited = {row[0] for row in rows}
+    return [doc for doc in docs if doc["id"] not in visited]
+
+
+if __name__ == "__main__":
+    args = parser.parse_args()
+    if args.reindex:
+        with sqlite3.connect("./visited.db") as conn:
+            c = conn.cursor()
+            c.execute("DELETE FROM indexed_documents")
+
+        logging.info("Fetching documents from Paperless-NGX")
+        ppngx = PaperlessNGXService()
+        docs = ppngx.get_data()
+        docs = filter_indexed_files(docs)
+        logging.info(f"Fetched {len(docs)} documents")
+
+        # Delete all chromadb data
+        ids = simba_docs.get(ids=None, limit=None, offset=0)
+        all_ids = ids["ids"]
+        if len(all_ids) > 0:
+            simba_docs.delete(ids=all_ids)
+
+        # Chunk documents
+        logging.info("Chunking documents now ...")
+        tag_lookup = ppngx.get_tags()
+        doctype_lookup = ppngx.get_doctypes()
+        chunk_data(docs, collection=simba_docs, doctypes=doctype_lookup)
+        logging.info("Done chunking documents")
+
+    # if args.index:
+    # with open(args.index) as file:
+    #     extension = args.index.split(".")[-1]
+    #     if extension == "pdf":
+    #         pdf_path = ppngx.download_pdf_from_id(id=document_id)
+    #         image_paths = pdf_to_image(filepath=pdf_path)
+    #         print(f"summarizing {file}")
+    #         generated_summary = summarize_pdf_image(filepaths=image_paths)
+    #     elif extension in [".md", ".txt"]:
+    #         chunk_text(texts=[file.readall()], collection=simba_docs)
+
+    if args.query:
+        logging.info("Consulting oracle ...")
+        print(
+            consult_oracle(
+                input=args.query,
+                collection=simba_docs,
+            )
+        )
+    else:
+        logging.info("please provide a query")

petmd_scrape_index.py

@@ -0,0 +1,24 @@
+from bs4 import BeautifulSoup
+import chromadb
+import httpx
+
+client = chromadb.PersistentClient(path="/Users/ryanchen/Programs/raggr/chromadb")
+
+# Scrape
+BASE_URL = "https://www.vet.cornell.edu"
+LIST_URL = "/departments-centers-and-institutes/cornell-feline-health-center/health-information/feline-health-topics"
+
+QUERY_URL = BASE_URL + LIST_URL
+r = httpx.get(QUERY_URL)
+soup = BeautifulSoup(r.text)
+
+container = soup.find("div", class_="field-body")
+a_s = container.find_all("a", href=True)
+
+new_texts = []
+
+for link in a_s:
+    endpoint = link["href"]
+    query_url = BASE_URL + endpoint
+    r2 = httpx.get(query_url)
+    article_soup = BeautifulSoup(r2.text)

pyproject.toml

@@ -4,4 +4,24 @@ version = "0.1.0"
 description = "Add your description here"
 readme = "README.md"
 requires-python = ">=3.13"
-dependencies = []
+dependencies = [
+    "chromadb>=1.1.0",
+    "python-dotenv>=1.0.0",
+    "flask>=3.1.2",
+    "httpx>=0.28.1",
+    "ollama>=0.6.0",
+    "openai>=2.0.1",
+    "pydantic>=2.11.9",
+    "pillow>=10.0.0",
+    "pymupdf>=1.24.0",
+    "black>=25.9.0",
+    "pillow-heif>=1.1.1",
+    "flask-jwt-extended>=4.7.1",
+    "bcrypt>=5.0.0",
+    "pony>=0.7.19",
+    "flask-login>=0.6.3",
+    "quart>=0.20.0",
+    "tortoise-orm>=0.25.1",
+    "quart-jwt-extended>=0.1.0",
+    "pre-commit>=4.3.0",
+]

query.py

@@ -0,0 +1,194 @@
+import json
+import os
+from typing import Literal
+import datetime
+from ollama import Client
+
+from openai import OpenAI
+
+from pydantic import BaseModel, Field
+
+# Configure ollama client with URL from environment or default to localhost
+ollama_client = Client(
+    host=os.getenv("OLLAMA_URL", "http://localhost:11434"), timeout=10.0
+)
+
+# This uses inferred filters — which means using LLM to create the metadata filters
+
+
+class FilterOperation(BaseModel):
+    op: Literal["$gt", "$gte", "$eq", "$ne", "$lt", "$lte", "$in", "$nin"]
+    value: str | list[str]
+
+
+class FilterQuery(BaseModel):
+    field_name: Literal["created_date, tags"]
+    op: FilterOperation
+
+
+class AndQuery(BaseModel):
+    op: Literal["$and", "$or"]
+    subqueries: list[FilterQuery]
+
+
+class GeneratedQuery(BaseModel):
+    fields: list[str]
+    extracted_metadata_fields: str
+
+
+class Time(BaseModel):
+    time: int
+
+
+DOCTYPE_OPTIONS = [
+    "Bill",
+    "Image Description",
+    "Insurance",
+    "Medical Record",
+    "Documentation",
+    "Letter",
+]
+
+
+class DocumentType(BaseModel):
+    type: list[str] = Field(description="type of document", enum=DOCTYPE_OPTIONS)
+
+
+PROMPT = """
+You are an information specialist that processes user queries. The current year is 2025. The user queries are all about
+a cat, Simba, and its records. The types of records are listed below. Using the query, extract the
+the date range the user is trying to query. You should return it as a JSON. The date tag is created_date. Return the date in epoch time.
+
+If the created_date cannot be ascertained, set it to epoch time start.
+
+You have several operators at your disposal:
+- $gt: greater than
+- $gte: greater than or equal
+- $eq: equal
+- $ne: not equal
+- $lt: less than
+- $lte: less than or equal to
+- $in: in
+- $nin: not in
+
+Logical operators:
+- $and, $or
+
+### Example 1
+Query: "Who is Simba's current vet?"
+Metadata fields: "{"created_date"}"
+Extracted metadata fields: {"created_date: {"$gt": "2025-01-01"}}
+
+### Example 2
+Query: "How many teeth has Simba had removed?"
+Metadata fields: {}
+Extracted metadata fields: {}
+
+### Example 3
+Query: "How many times has Simba been to the vet this year?"
+Metadata fields: {"created_date"}
+Extracted metadata fields: {"created_date": {"gt": "2025-01-01"}}
+
+document_types:
+- aftercare
+- bill
+- insurance claim
+- medical records
+
+Only return the extracted metadata fields. Make sure the extracted metadata fields are valid JSON
+"""
+
+
+DOCTYPE_PROMPT = f"""You are an information specialist that processes user queries. A query can have two tags attached from the following options. Based on the query, determine which of the following options is most appropriate: {",".join(DOCTYPE_OPTIONS)}
+
+### Example 1
+Query: "Who is Simba's current vet?"
+Tags: ["Bill", "Medical Record"]
+
+
+### Example 2
+Query: "Who does Simba know?"
+Tags: ["Letter", "Documentation"]
+"""
+
+
+class QueryGenerator:
+    def __init__(self) -> None:
+        pass
+
+    def date_to_epoch(self, date_str: str) -> float:
+        split_date = date_str.split("-")
+        date = datetime.datetime(
+            int(split_date[0]),
+            int(split_date[1]),
+            int(split_date[2]),
+            0,
+            0,
+            0,
+        )
+
+        return date.timestamp()
+
+    def get_doctype_query(self, input: str):
+        client = OpenAI()
+        response = client.chat.completions.create(
+            messages=[
+                {
+                    "role": "system",
+                    "content": "You are an information specialist that is really good at deciding what tags a query should have",
+                },
+                {"role": "user", "content": DOCTYPE_PROMPT + " " + input},
+            ],
+            model="gpt-4o",
+            response_format={
+                "type": "json_schema",
+                "json_schema": {
+                    "name": "document_type",
+                    "schema": DocumentType.model_json_schema(),
+                },
+            },
+        )
+
+        response_json_str = response.choices[0].message.content
+        type_data = json.loads(response_json_str)
+        metadata_query = {"document_type": {"$in": type_data["type"]}}
+        return metadata_query
+
+    def get_query(self, input: str):
+        client = OpenAI()
+        response = client.responses.parse(
+            model="gpt-4o",
+            input=[
+                {"role": "system", "content": PROMPT},
+                {"role": "user", "content": input},
+            ],
+            text_format=GeneratedQuery,
+        )
+        print(response.output)
+        query = json.loads(response.output_parsed.extracted_metadata_fields)
+        # response: ChatResponse = ollama_client.chat(
+        # model="gemma3n:e4b",
+        # messages=[
+        # {"role": "system", "content": PROMPT},
+        # {"role": "user", "content": input},
+        # ],
+        # format=GeneratedQuery.model_json_schema(),
+        # )
+
+        # query = json.loads(
+        # json.loads(response["message"]["content"])["extracted_metadata_fields"]
+        # )
+        # date_key = list(query["created_date"].keys())[0]
+        # query["created_date"][date_key] = self.date_to_epoch(
+        # query["created_date"][date_key]
+        # )
+
+        # if "$" not in date_key:
+        # query["created_date"]["$" + date_key] = query["created_date"][date_key]
+
+        return query
+
+
+if __name__ == "__main__":
+    qg = QueryGenerator()
+    print(qg.get_doctype_query("How heavy is Simba?"))

raggr-frontend/.gitignore

@@ -0,0 +1,16 @@
+# Local
+.DS_Store
+*.local
+*.log*
+
+# Dist
+node_modules
+dist/
+
+# Profile
+.rspack-profile-*/
+
+# IDE
+.vscode/*
+!.vscode/extensions.json
+.idea

raggr-frontend/README.md

@@ -0,0 +1,36 @@
+# Rsbuild project
+
+## Setup
+
+Install the dependencies:
+
+```bash
+pnpm install
+```
+
+## Get started
+
+Start the dev server, and the app will be available at [http://localhost:3000](http://localhost:3000).
+
+```bash
+pnpm dev
+```
+
+Build the app for production:
+
+```bash
+pnpm build
+```
+
+Preview the production build locally:
+
+```bash
+pnpm preview
+```
+
+## Learn more
+
+To learn more about Rsbuild, check out the following resources:
+
+- [Rsbuild documentation](https://rsbuild.rs) - explore Rsbuild features and APIs.
+- [Rsbuild GitHub repository](https://github.com/web-infra-dev/rsbuild) - your feedback and contributions are welcome!

raggr-frontend/package.json

@@ -0,0 +1,26 @@
+{
+  "name": "raggr-frontend",
+  "version": "1.0.0",
+  "private": true,
+  "type": "module",
+  "scripts": {
+    "build": "rsbuild build",
+    "dev": "rsbuild dev --open",
+    "preview": "rsbuild preview"
+  },
+  "dependencies": {
+    "axios": "^1.12.2",
+    "marked": "^16.3.0",
+    "react": "^19.1.1",
+    "react-dom": "^19.1.1",
+    "react-markdown": "^10.1.0"
+  },
+  "devDependencies": {
+    "@rsbuild/core": "^1.5.6",
+    "@rsbuild/plugin-react": "^1.4.0",
+    "@tailwindcss/postcss": "^4.0.0",
+    "@types/react": "^19.1.13",
+    "@types/react-dom": "^19.1.9",
+    "typescript": "^5.9.2"
+  }
+}

raggr-frontend/postcss.config.mjs

@@ -0,0 +1,5 @@
+export default {
+	plugins: {
+		"@tailwindcss/postcss": {},
+	},
+};

raggr-frontend/rsbuild.config.ts

@@ -0,0 +1,6 @@
+import { defineConfig } from '@rsbuild/core';
+import { pluginReact } from '@rsbuild/plugin-react';
+
+export default defineConfig({
+  plugins: [pluginReact()],
+});

raggr-frontend/src/App.css

@@ -0,0 +1,6 @@
+@import "tailwindcss";
+
+body {
+	margin: 0;
+	font-family: Inter, Avenir, Helvetica, Arial, sans-serif;
+}

raggr-frontend/src/App.tsx

@@ -0,0 +1,204 @@
+import { useEffect, useState } from "react";
+import axios from "axios";
+import ReactMarkdown from "react-markdown";
+
+import "./App.css";
+
+type QuestionAnswer = {
+  question: string;
+  answer: string;
+};
+
+type QuestionBubbleProps = {
+  text: string;
+};
+
+type AnswerBubbleProps = {
+  text: string;
+  loading: string;
+};
+
+type QuestionAnswerPairProps = {
+  question: string;
+  answer: string;
+  loading: boolean;
+};
+
+type Conversation = {
+  title: string;
+  id: string;
+};
+
+type Message = {
+  text: string;
+  speaker: "simba" | "user";
+};
+
+type ConversationMenuProps = {
+  conversations: Conversation[];
+};
+
+const ConversationMenu = ({ conversations }: ConversationMenuProps) => {
+  return (
+    <div className="absolute bg-white w-md rounded-md shadow-xl m-4 p-4">
+      <p className="py-2 px-4 rounded-md w-full text-xl font-bold">askSimba!</p>
+      {conversations.map((conversation) => (
+        <p className="py-2 px-4 rounded-md hover:bg-stone-200 w-full text-xl font-bold cursor-pointer">
+          {conversation.title}
+        </p>
+      ))}
+    </div>
+  );
+};
+
+const QuestionBubble = ({ text }: QuestionBubbleProps) => {
+  return <div className="rounded-md bg-stone-200 p-3">🤦: {text}</div>;
+};
+
+const AnswerBubble = ({ text, loading }: AnswerBubbleProps) => {
+  return (
+    <div className="rounded-md bg-orange-100 p-3">
+      {loading ? (
+        <div className="flex flex-col w-full animate-pulse gap-2">
+          <div className="flex flex-row gap-2 w-full">
+            <div className="bg-gray-400 w-1/2 p-3 rounded-lg" />
+            <div className="bg-gray-400 w-1/2 p-3 rounded-lg" />
+          </div>
+          <div className="flex flex-row gap-2 w-full">
+            <div className="bg-gray-400 w-1/3 p-3 rounded-lg" />
+            <div className="bg-gray-400 w-2/3 p-3 rounded-lg" />
+          </div>
+        </div>
+      ) : (
+        <div className="flex flex-col">
+          <ReactMarkdown>{"🐈: " + text}</ReactMarkdown>
+        </div>
+      )}
+    </div>
+  );
+};
+
+const QuestionAnswerPair = ({
+  question,
+  answer,
+  loading,
+}: QuestionAnswerPairProps) => {
+  return (
+    <div className="flex flex-col gap-4">
+      <QuestionBubble text={question} />
+      <AnswerBubble text={answer} loading={loading} />
+    </div>
+  );
+};
+
+const App = () => {
+  const [query, setQuery] = useState<string>("");
+  const [answer, setAnswer] = useState<string>("");
+  const [simbaMode, setSimbaMode] = useState<boolean>(false);
+  const [questionsAnswers, setQuestionsAnswers] = useState<QuestionAnswer[]>(
+    [],
+  );
+  const [messages, setMessages] = useState<Message[]>([]);
+  const [conversations, setConversations] = useState<Conversation[]>([
+    { title: "simba meow meow", id: "uuid" },
+  ]);
+
+  const simbaAnswers = ["meow.", "hiss...", "purrrrrr", "yowOWROWWowowr"];
+
+  useEffect(() => {
+    axios.get("/api/messages").then((result) => {
+      setMessages(
+        result.data.messages.map((message) => {
+          return {
+            text: message.text,
+            speaker: message.speaker,
+          };
+        }),
+      );
+    });
+  }, []);
+
+  const handleQuestionSubmit = () => {
+    let currMessages = messages.concat([{ text: query, speaker: "user" }]);
+    setMessages(currMessages);
+    if (simbaMode) {
+      console.log("simba mode activated");
+      const randomIndex = Math.floor(Math.random() * simbaAnswers.length);
+      const randomElement = simbaAnswers[randomIndex];
+      setAnswer(randomElement);
+      setQuestionsAnswers(
+        questionsAnswers.concat([
+          {
+            question: query,
+            answer: randomElement,
+          },
+        ]),
+      );
+      return;
+    }
+    const payload = { query: query };
+    axios.post("/api/query", payload).then((result) => {
+      setQuestionsAnswers(
+        questionsAnswers.concat([
+          { question: query, answer: result.data.response },
+        ]),
+      );
+      setMessages(
+        currMessages.concat([{ text: result.data.response, speaker: "simba" }]),
+      );
+    });
+  };
+  const handleQueryChange = (event) => {
+    setQuery(event.target.value);
+  };
+  return (
+    <div className="h-screen bg-opacity-20">
+      <div className="bg-white/85 h-screen">
+        <div className="flex flex-row justify-center py-4">
+          <div className="flex flex-col gap-4 min-w-xl max-w-xl">
+            <header className="flex flex-row justify-center gap-2 grow sticky top-0 z-10 bg-white">
+              <h1 className="text-3xl">ask simba!</h1>
+            </header>
+            {/*{questionsAnswers.map((qa) => (
+              <QuestionAnswerPair question={qa.question} answer={qa.answer} />
+            ))}*/}
+            {messages.map((msg) => {
+              if (msg.speaker == "simba") {
+                return <AnswerBubble text={msg.text} loading="" />;
+              }
+
+              return <QuestionBubble text={msg.text} />;
+            })}
+            <footer className="flex flex-col gap-2 sticky bottom-0">
+              <div className="flex flex-row justify-between gap-2 grow">
+                <textarea
+                  type="text"
+                  className="p-4 border border-blue-200 rounded-md grow bg-white"
+                  onChange={handleQueryChange}
+                />
+              </div>
+              <div className="flex flex-row justify-between gap-2 grow">
+                <button
+                  className="p-4 border border-blue-400 bg-blue-200 hover:bg-blue-400 cursor-pointer rounded-md flex-grow"
+                  onClick={() => handleQuestionSubmit()}
+                  type="submit"
+                >
+                  Submit
+                </button>
+              </div>
+              <div className="flex flex-row justify-center gap-2 grow">
+                <input
+                  type="checkbox"
+                  onChange={(event) => setSimbaMode(event.target.checked)}
+                />
+                <p>simba mode?</p>
+              </div>
+            </footer>
+          </div>
+        </div>
+      </div>
+    </div>
+  );
+};
+
+export default App;

raggr-frontend/src/env.d.ts

@@ -0,0 +1,11 @@
+/// <reference types="@rsbuild/core/types" />
+
+/**
+ * Imports the SVG file as a React component.
+ * @requires [@rsbuild/plugin-svgr](https://npmjs.com/package/@rsbuild/plugin-svgr)
+ */
+declare module '*.svg?react' {
+  import type React from 'react';
+  const ReactComponent: React.FunctionComponent<React.SVGProps<SVGSVGElement>>;
+  export default ReactComponent;
+}

raggr-frontend/src/index.tsx

@@ -0,0 +1,13 @@
+import React from 'react';
+import ReactDOM from 'react-dom/client';
+import App from './App';
+
+const rootEl = document.getElementById('root');
+if (rootEl) {
+  const root = ReactDOM.createRoot(rootEl);
+  root.render(
+    <React.StrictMode>
+      <App />
+    </React.StrictMode>,
+  );
+}

raggr-frontend/tsconfig.json

@@ -0,0 +1,25 @@
+{
+  "compilerOptions": {
+    "lib": ["DOM", "ES2020"],
+    "jsx": "react-jsx",
+    "target": "ES2020",
+    "noEmit": true,
+    "skipLibCheck": true,
+    "useDefineForClassFields": true,
+
+    /* modules */
+    "module": "ESNext",
+    "moduleDetection": "force",
+    "moduleResolution": "bundler",
+    "verbatimModuleSyntax": true,
+    "resolveJsonModule": true,
+    "allowImportingTsExtensions": true,
+    "noUncheckedSideEffectImports": true,
+
+    /* type checking */
+    "strict": true,
+    "noUnusedLocals": true,
+    "noUnusedParameters": true
+  },
+  "include": ["src"]
+}

request.py

@@ -1,19 +1,86 @@
 import os
+import tempfile
 import httpx
+import logging
 
 from dotenv import load_dotenv
 
 load_dotenv()
 
+logging.basicConfig(level=logging.INFO)
+
 
 class PaperlessNGXService:
     def __init__(self):
         self.base_url = os.getenv("BASE_URL")
         self.token = os.getenv("PAPERLESS_TOKEN")
-        self.url = f"http://{os.getenv("BASE_URL")}/api/documents/?query=simba"
-        self.headers = {"Authorization": f"Token {os.getenv("PAPERLESS_TOKEN")}"}
+        self.url = f"http://{os.getenv('BASE_URL')}/api/documents/?tags__id=8"
+        self.headers = {"Authorization": f"Token {os.getenv('PAPERLESS_TOKEN')}"}
 
     def get_data(self):
         print(f"Getting data from: {self.url}")
         r = httpx.get(self.url, headers=self.headers)
-        return r.json()["results"]
+        results = r.json()["results"]
+
+        nextLink = r.json().get("next")
+
+        while nextLink:
+            r = httpx.get(nextLink, headers=self.headers)
+            results += r.json()["results"]
+            nextLink = r.json().get("next")
+
+        return results
+
+    def get_doc_by_id(self, doc_id: int):
+        url = f"http://{os.getenv('BASE_URL')}/api/documents/{doc_id}/"
+        r = httpx.get(url, headers=self.headers)
+        return r.json()
+
+    def download_pdf_from_id(self, id: int) -> str:
+        download_url = f"http://{os.getenv('BASE_URL')}/api/documents/{id}/download/"
+        response = httpx.get(
+            download_url, headers=self.headers, follow_redirects=True, timeout=30
+        )
+        response.raise_for_status()
+        # Use a temporary file for the downloaded PDF
+        temp_file = tempfile.NamedTemporaryFile(delete=False, suffix=".pdf")
+        temp_file.write(response.content)
+        temp_file.close()
+        temp_pdf_path = temp_file.name
+        pdf_to_process = temp_pdf_path
+        return pdf_to_process
+
+    def upload_cleaned_content(self, document_id, data):
+        PUTS_URL = f"http://{os.getenv('BASE_URL')}/api/documents/{document_id}/"
+        r = httpx.put(PUTS_URL, headers=self.headers, data=data)
+        r.raise_for_status()
+
+    def upload_description(self, description_filepath, file, title, exif_date: str):
+        POST_URL = f"http://{os.getenv('BASE_URL')}/api/documents/post_document/"
+        files = {"document": ("description_filepath", file, "application/txt")}
+        data = {
+            "title": title,
+            "create": exif_date,
+            "document_type": 3,
+            "tags": [7],
+        }
+
+        r = httpx.post(POST_URL, headers=self.headers, data=data, files=files)
+        r.raise_for_status()
+
+    def get_tags(self):
+        GET_URL = f"http://{os.getenv('BASE_URL')}/api/tags/"
+        r = httpx.get(GET_URL, headers=self.headers)
+        data = r.json()
+        return {tag["id"]: tag["name"] for tag in data["results"]}
+
+    def get_doctypes(self):
+        GET_URL = f"http://{os.getenv('BASE_URL')}/api/document_types/"
+        r = httpx.get(GET_URL, headers=self.headers)
+        data = r.json()
+        return {doctype["id"]: doctype["name"] for doctype in data["results"]}
+
+
+if __name__ == "__main__":
+    pp = PaperlessNGXService()
+    pp.get_data()

startup.sh

@@ -0,0 +1,7 @@
+#!/bin/bash
+
+echo "Starting reindex process..."
+python main.py "" --reindex
+
+echo "Starting Flask application..."
+python app.py