QuantAnalyst AI
An end-to-end stock analysis platform: market data, financial news sentiment, rule-based technical scoring and an LLM-written report, served through a hypermedia interface with no custom JavaScript.
- Role
- Software developer
- Built with
- Python, Django, HTMX, PyTorch, FinBERT, LangGraph, Plotly, SQLite
What it does
Enter a ticker and the pipeline fetches daily market data, computes technical features (moving average, returns, volatility, RSI), scores recent news with FinBERT, and turns it into a bullish-probability score. A LangGraph agent then asks a large language model for a Markdown report. Chart, score, sentiment and report are swapped into the page by HTMX.
Engineering decisions
The server renders all HTML, and the API returns partial HTML. The app calls the LLM through an OpenAI-compatible gateway, so environment variables choose the model. FinBERT loads once and stays in memory.
Two caches cut costs. API responses are cached on disk, and a repeat analysis of a ticker on the same day comes from the database without re-running the ML and LLM steps. The price-movement classifier is still rule-based; a trained model is planned.