NeuraPDF is a full-stack application that allows users to upload and embed PDFs, then chat with them using AI-powered natural language queries. It integrates Flask (backend) and Svelte (frontend), with Redis and Celery for asynchronous task processing.

Flask backend application.

• celery/: Celery integration with Flask. Includes:

  • __init__.py: Initializes Celery with Flask context.
  • worker.py: Entry point to launch the Celery worker process.

• chat/: Core logic for PDF chat functionality.

  • chat.py: Handles query-to-response logic.
  • create_embeddings.py: Generates vector embeddings for PDF content.
  • score.py: Scores responses based on relevance.
  • models/: Placeholder for chat-related data models.

• web/: Flask web layer and APIs.

  • api.py: Defines main Flask API routes.
  • config/: Configuration initialization.
  • db/: SQLAlchemy models for:
    • conversation.py, message.py, pdf.py, user.py: Represents app data.
  • files.py: PDF file processing utilities.
  • hooks.py: App startup/shutdown hooks.
  • tasks/: Celery background jobs (e.g. embedding generation).
  • views/: Flask route handlers for user-facing endpoints.

Frontend written in Svelte.

• src/api/axios.ts: Axios client config for API calls.
• components/: Reusable UI components like chat interface, navbar, forms.
• routes/: SvelteKit routing.
  • (app)/auth/: Auth pages (signin, signup, signout).
  • (app)/chat/: Main chat interface.
  • (app)/documents/: Upload and view documents.
  • (app)/scores/: View scoring results.
• store/: Svelte stores for reactive state management (auth, chat, errors, etc.).
• static/: Static assets like favicon, animation gif, and OpenAPI docs.
• Svelte + Tailwind + TypeScript configuration files.

SQLite database file (auto-created for local development).

• Pipfile / requirements.txt: Python dependencies.
• tasks.py: Entry point or glue code for CLI or debugging utilities.
• README.md: Project documentation.

• Backend: Flask, Celery, SQLAlchemy, Redis
• Frontend: SvelteKit, Tailwind CSS, TypeScript
• AI / NLP: Embedding models for vector search
• Database: SQLite (dev), pluggable for production

pipenv install
pipenv shell
flask --app app.web init-db

The app requires three separate processes:

1. Python server
2. Celery worker
3. Redis server

pipenv shell
inv dev

pipenv shell
inv devworker

redis-server

pipenv shell
flask --app app.web init-db

==> use langchain-openai=0.0.8

 pip install langchain-openai==0.0.8

==> downgrade pydantic<2.0.0

 pip install "pydantic<2.0.0"

NeuraPDF enables users to upload PDFs, generate embeddings, and interact with them through a chat interface. It processes heavy embedding tasks asynchronously with Celery and Redis and provides a polished UI using SvelteKit.