Project #1

pondit_moshAI

Laravel + Python document RAG application with:

Authenticated users private document storage queued PDF ingestion
Qdrant-backed retrieval filtered by user_id Ollama chat with streamed responses persisted chat sessions with UUID-based URLs

The project is split into two parts:

Laravel app:
- authentication
- dashboard and document management
- private file serving and download
- queued ingestion trigger
- chat UI and streamed chat responses
Python pipeline in scripts/:
- PDF loading
- chunking
- Hugging Face embeddings using all-MiniLM-L6-v2
- Qdrant retrieval and storage

Current RAG flow:

A logged-in user uploads a PDF from the dashboard.
Laravel stores the original file in private storage and records it in the documents table.
A queued job copies the file into scripts/data/<category>/ and runs the Python ingest command with the current user_id.
The Python script embeds chunks and stores them in Qdrant with metadata.user_id and metadata.category.
In chat, Laravel asks the Python script for top matching chunks for the current user.
Laravel injects those chunks into a system prompt and streams the final answer from Ollama.

Main Features

Document uploads are stored for reference. Originals are not deleted after ingestion.
Documents are tracked in the database with ingestion status.
Users can only read and download their own stored files.
Categories are centrally maintained in config/category.php.
Chat sessions are persisted in chat_sessions and chat_messages.
Chat session URLs use UUIDs, for example /chat?session=<uuid>.
Last 20 messages are loaded into model memory per chat session.
Chat uses normal streamed HTTP responses. No WebSocket server is required.

Requirements