What it is

Fidash is a self-hosted financial research platform that aggregates stock news, runs sentiment analysis, and executes automated trading strategies on prediction markets. It runs on a Raspberry Pi as a production server, exposed publicly through a Cloudflare tunnel, and is operated entirely through a Streamlit dashboard that communicates with a FastAPI backend over HTTP.

Architecture

The architecture follows a layered pattern: endpoints -> controllers -> services -> repositories -> SQLAlchemy models. A FastAPI app factory wires together over a dozen domain routers covering tickers, news, sentiment, scans, backtests, alerts, watchlists, paper trading, and a Kalshi prediction-market bot. An APScheduler instance runs background jobs for data ingestion from Polygon, NewsAPI, AlphaVantage, Alpaca, and RSS feeds, alongside alert evaluation, queued workflow execution, and scheduled Kalshi reconciliation. The Streamlit frontend uses aiohttp to call the backend asynchronously, with a centralized theme system (dashboard_theme.py) that owns all colors and chart defaults.

Technical highlights

The stack is Python 3.11 throughout. Sentiment scoring uses VADER and onnxruntime-served models. The scanner generates options strategy cards (put_credit_spread, iron_condor, short_strangle, short_straddle) from scan results. The Kalshi bot subsystem includes discrete trading bots for weather, gas prices, jobless claims, oil, and S&P 500 markets — each following a data_loader -> model -> bot pattern with walk-forward backtesting, paper and live modes, position limits, and cross-restart deduplication via JSONL and SQLite persistence. The database is PostgreSQL in production with Alembic migrations, SQLite for local development. Monitoring is provided by Prometheus and Grafana via Docker Compose.

Where it’s at

The project is in active development; the weather bot runs live on Kalshi and several other bots operate in paper mode.