WatAIOliver

Note: A FastAPI backend is being added in the backend/ folder. The chat logic from Oliver-Legacy is being migrated here to support real Qwen model chat functionality.

This is the Oliver repo for WatAI — an LLM-based RAG project. Oliver is an intelligent assistant who helps instructors and students interact with course materials through conversational AI. This implementation is a refresh of the legacy Oliver project.

Instructors can upload course content, and Oliver uses a RAG pipeline to generate context-aware answers when students ask questions. The first scheduled course for deployment is MTE 182 by Albert Jiang in the Fall term.

WatAIOliver combines a React frontend with a FastAPI backend and a modern RAG system. Users can upload material and chat with the assistant.

Repository Layout

• frontend/ – React application built with Vite and TailwindCSS
• backend/ – FastAPI service exposing chat and file APIs
• machine_learning/ – Embedding and RAG services, plus optional agent system

Quick Start

1. Install dependencies

   On macOS/Linux you can run the helper script:

   ./setup.sh

   remember to run chmod +x setup.sh before ./ it.

./setup.sh          # setup + run everything (default)
./setup.sh setup    # setup only
./setup.sh start    # run services
./setup.sh stop     # stop services
./setup.sh status   # status

---

Or install manually:

pip install -r backend/requirements.txt
pip install -r machine_learning/pdf_processor/requirements.txt
pip install -r machine_learning/rag_system/requirements.txt -c constraints.txt --upgrade --upgrade-strategy eager
pip install -r machine_learning/ai_agents/requirements.txt

Windows users should follow windows_proj_env_setup.md. 3. Environment files – copy the provided examples to .env inside backend/ and machine_learning/rag_system/ and update credentials. 4. Run services

# Backend
PYTHONPATH=backend python -m uvicorn backend.main:app --reload --host 0.0.0.0 --port 8000
# Frontend
cd frontend && npm ci
npm run dev

Optional services:

# PDF processor
PYTHONPATH=machine_learning python -m uvicorn pdf_processor.main:app --reload --host 0.0.0.0 --port 8001
# RAG system
PYTHONPATH=machine_learning python -m uvicorn rag_system.app.main:app --reload --host 0.0.0.0 --port 8002
# Agent system
PYTHONPATH=machine_learning python -m uvicorn ai_agents.app.main:app --reload --host 0.0.0.0 --port 8003

Logging

The backend uses a rotating file logger located at backend/src/app.log. All requests are logged with their status code and processing time, making it easy to trace errors.

Features

• Chat with persistent conversations
• File uploads (PDF and text)
• KaTeX formula rendering
• Retrieval‑augmented responses using the RAG system
• Optional multi‑agent debate mode

Building Frontend

To create a production build of the frontend:

cd frontend
npm run build

The output is placed in frontend/dist.

Technology Stack

• Frontend: React, TailwindCSS
• Backend: FastAPI
• LLM: Newer LLM APIs (replacing legacy APIs)
• Database: Supabase (PostgreSQL)
• Vector DB: ChromaDB or pgvector (choose based on team preference)
• CI/CD: GitHub Actions (or deploy scripts for separate parts)

Git Workflow

We maintain three main branches:

• main: production branch
• staging: pre-production
• development: active development

Branching rules:

• Only staging can be merged into main.
• Only development can be merged into staging.
• Create feature branches from development, e.g. @frontend/login or @backend/chat-api.
• Make a PR when a feature is implemented for easier review.

Let's build an awesome new Oliver together!