Skip to content

Repository files navigation

Themis Logo

Themis

Automated exploration and auditing of suspicious public procurement in Chile

themis.lat


What is Themis?

Themis is an AI-powered platform that automatically analyzes public tenders from Chile's Mercado Público to detect:

  • Regulatory violations - Non-compliance with procurement laws and regulations
  • Suspicious patterns - Anomalies in award decisions and bidding behavior
  • Procurement irregularities - Red flags in purchasing processes

Themis brings transparency to public spending by making it easy to investigate and audit government contracts.

How it works

Themis uses a multi-agent AI system built with LangGraph that can:

  1. Fetch tender data from Mercado Público's API
  2. Analyze documents including contracts, specifications, and award decisions
  3. Detect anomalies using specialized fraud detection agents
  4. Generate reports summarizing findings and ranking issues by severity

Tech Stack

Frontend

  • React 19 + TypeScript + Vite
  • Mosaic for data visualization
  • Embedding Atlas for document exploration

Backend

  • FastAPI + Python 3.13
  • LangGraph + LangChain for AI orchestration
  • PostgreSQL + SQLAlchemy
  • Mistral AI for language processing

Infrastructure

  • Docker & Docker Compose

Getting Started

Prerequisites

  • Docker Desktop

Environment Variables

Create a backend/.env file:

DATABASE_URL=postgresql://postgres:postgres@postgres:5432/postgres

ANTHROPIC_API_KEY=your-anthropic-api-key
MISTRAL_API_KEY=your-mistral-api-key
OPENROUTER_API_KEY=your-openrouter-api-key

LANGSMITH_TRACING=true
LANGSMITH_ENDPOINT=https://api.smith.langchain.com
LANGSMITH_API_KEY=your-langsmith-api-key
LANGSMITH_PROJECT=your-project-name

ADMIN_API_KEY=your-admin-key
DISCORD_WEBHOOK_URL=your-discord-webhook-url

Run

docker compose up

Team

Built by:

License

This project was created for the Platanus Hackathon 2025 and is open source.

About

No description, website, or topics provided.

Resources

Stars

1 star

Watchers

0 watching

Forks

Releases

Packages

Contributors

Languages