FAIN, AI Trading Ecosystem

FAIN is a trading terminal for the Tunis stock exchange, built at the IHEC CodeLab hackathon. It tracks all 82 listed companies live, forecasts short-term prices, builds a portfolio and explains its choices, reads the mood from Tunisian financial news, and answers questions about market rules through a chatbot.

The Tunis market has almost no modern tooling. Investors work off fragmented, manual processes with no real quantitative help, and nothing brings live prices, forecasting, explained portfolio advice, and local sentiment together in one place.

1. 01

   One gateway, many services

   A FastAPI gateway routes to a dozen small services, each doing one job: prices, market data, forecasting, anomaly detection, sentiment, portfolio, notifications, and more.

2. 02

   Short-term price forecasts

   An XGBoost model trains on the fly for each request, using more than 30 technical indicators, and returns a five-day forecast with confidence bands that widen the further out it looks.

3. 03

   A portfolio that explains itself

   A reinforcement-learning agent allocates the portfolio using live macro data from the World Bank, the IMF, and the Tunisian central bank. SHAP then shows which factors drove each choice, and an LLM writes the reasoning up in plain French.

4. 04

   Market mood from local news

   It scrapes Tunisian financial outlets and scores sentiment per company, with live social-media checks layered on top.

5. 05

   A chatbot for the rulebook

   A retrieval chatbot answers questions over the official exchange and regulator documents, keeping context across a conversation.

A Next.js frontend talks only to a FastAPI gateway. The gateway reads prices straight from Postgres and proxies everything else to the downstream services: forecasting, anomaly detection, sentiment, portfolio, auth, and notifications. A separate jobs service runs the scheduled work, like the market pulse and the daily report.

1. 01

   Choosing XGBoost over LSTM for forecasting was the right hackathon call, on-the-fly training per request with early stopping converges in seconds, while LSTM fine-tuning per stock would take hours. Directional accuracy on thin BVMT data was comparable.

2. 02

   Using PPO Reinforcement Learning for portfolio allocation instead of classical Markowitz lets the agent learn non-linear risk/return tradeoffs and directly incorporate macro signals (BCT policy rate, IMF debt/GDP) into the reward function, something mean-variance optimization cannot do.

3. 03

   The two-layer sentiment approach (Gemini Flash Lite for scraped archives + Perplexity Sonar for real-time social) proved essential: static scrapers lag by hours on breaking news, while Sonar returns live discussion threads within seconds.

4. 04

   SHAP on top of the PPO agent's feature set was the single biggest trust-builder in the demo, once judges could see that 'BCT policy rate' and 'BIAT volatility' were the top drivers of a recommendation, skepticism about the RL black box evaporated instantly.

5. 05

   The API Gateway pattern was the most important architectural decision for team velocity, each engineer owned one service on its own port, and the gateway unified everything behind a single URL without ever touching the frontend routing code.