EADA — Enterprise Autonomous Data Analyst

A multi-agent AI platform that autonomously analyses data files and documents using a self-correcting agent graph, built entirely with free tools on a CPU-only laptop.

Features

• Multi-agent architecture — Router, Planner, Analyst, RAG, Critic, and Summarizer agents collaborate via LangGraph
• Data file analysis — Upload CSV, Excel, JSON, or Parquet files and query them in natural language
• Document Q&A — Ingest PDF, DOCX, TXT files and ask questions via RAG pipeline
• Self-correction — Critic agent scores responses and retries up to 2 times if quality is low
• Conversation memory — Redis-backed memory persists context across turns
• Observability — Full tracing via Langfuse, structured JSON logging via structlog
• Production-ready — Dockerized stack with Nginx reverse proxy, healthchecks, and GHCR CI/CD

Stack

Layer	Technology
LLM	Gemini 2.5 Flash (primary), Groq LLaMA 3.3 70B (fallback)
Agent Framework	LangGraph 1.2
Backend	FastAPI + uvicorn
Database	PostgreSQL 16 (asyncpg + SQLAlchemy)
Vector Store	Qdrant 1.13
Cache / Memory	Redis 7
Observability	Langfuse + structlog
Frontend	Vite + React + TypeScript + Tailwind CSS
Infra	Docker Compose + Nginx
CI/CD	GitHub Actions + GHCR

Quick Start

Prerequisites

• Docker Desktop
• A Gemini API key from https://aistudio.google.com/apikey

1. Clone the repo

git clone https://github.com/codewithleo1/eada.git cd eada

2. Configure environment

cp .env.example .env

Edit .env and add your GEMINI_API_KEY

3. Start the full stack

docker compose -f docker-compose.prod.yml up --build -d

4. Open the app

Visit http://localhost — register a new account on first run.

Development Setup (Windows)

`powershell $env:Path = "C:\Users<you>.local\bin;$env:Path" docker compose up -d uv run uvicorn backend.main:app --reload

In a separate terminal:

cd frontend npm run dev `powershell

Running Tests

uv run pytest backend/tests/ -v

68 unit tests covering agent graph routing, tool execution, LLM gateway, Redis memory, and response scoring.

Project Structure

backend/agents/ — LangGraph agent nodes backend/api/routes/ — FastAPI endpoints backend/db/ — SQLAlchemy models and Alembic migrations backend/llm/ — LiteLLM gateway with fallback backend/memory/ — Redis-backed agent memory backend/rag/ — Chunker, embedder, Qdrant vector store backend/tools/ — DuckDB SQL tool, file schema tool backend/evaluation/ — LLM-based response scorer frontend/src/ — Vite + React + TypeScript chat UI infra/ — Nginx config, Postgres init script

License

MIT