AIRA – AI Research Assistant

AIRA is a full-stack app that turns your documents into a searchable, chat-based research assistant. Upload PDFs and text files, then ask questions in natural language. A FastAPI backend handles ingestion, embeddings, and RAG; a React frontend provides the UI; and Ollama runs the LLM locally.

Project structure

Directory	Stack	Role
frontend/	React, TypeScript, Vite	Document management, model selector, chat UI
backend/	FastAPI, SQLite, ChromaDB	API, document processing, embeddings, RAG, chat sessions

Features

• Document management – Upload, list, and delete documents (PDF, .txt, .md). Text is extracted, chunked, embedded, and stored for semantic search.
• RAG search – Query your documents with natural language; the API returns relevant chunks and uses them as context for answers.
• Chat – Multi-turn chat with streaming responses. Sessions are stored in SQLite and can be resumed; history is persisted in the browser.
• Model selection – Pick which Ollama model to use for chat from the UI or via the API.
• API docs – OpenAPI/Swagger at /docs when the backend is running.

Prerequisites

• Node.js 18+
• Python 3.11+
• Ollama – Install and pull a model
• npm (or yarn/pnpm)

Quick start

1. Ollama

Install Ollama and pull a model:

ollama pull qwen3:8b

2. Backend

From the project root:

cd backend
python -m venv venv
source venv/bin/activate   # Windows: venv\Scripts\activate
pip install -r requirements.txt
uvicorn main:app --reload --port 8000

API: http://localhost:8000

3. Frontend

In another terminal:

cd frontend
npm install
npm run dev

App: http://localhost:5173. The dev server proxies /api to the backend.

API overview

When the backend is running, interactive docs: http://localhost:8000/docs

Endpoint	Method	Description
/health	GET	Health check
/api/ollama/status	GET	Ollama connection and model list
/api/models	GET	List available Ollama models
/api/model/select	POST	Set the model used for chat (body: {"model_name": "qwen3:8b"})
/api/documents	GET	List uploaded documents
/api/documents/upload	POST	Upload a document (PDF, .txt, .md)
/api/documents/{id}	DELETE	Delete a document and its vectors
/api/search	GET	Semantic search (q, top_k, optional doc_ids)
/api/chat/session/new	POST	Create a chat session
/api/chat/session/{id}	GET / DELETE	Get or delete a session
/api/chat/message	POST	Send a message (streaming SSE response)

How it works

1. Upload – You upload a file; the backend extracts text, splits it into chunks, generates embeddings (Ollama nomic-embed-text), and stores them in ChromaDB.
2. Search – A query is embedded and compared to stored chunks; the most relevant chunks are returned (and optionally passed to the LLM).
3. Chat – Each message triggers retrieval over your documents; the retrieved context plus the conversation is sent to the selected Ollama model, and the reply is streamed back via SSE. Sessions and messages are stored in SQLite.

Development

• Backend: uvicorn main:app --reload (run from backend/).
• Frontend: npm run dev (run from frontend/).

Both support hot reload.

Production build (frontend)

cd frontend
npm run build

Output is in frontend/dist/. Serve it with any static host; ensure requests to /api are proxied to the FastAPI backend.

License

MIT