Faction: Stock Analysis App
A desktop app for evaluating stocks through technical and fundamental analysis, powered by on-device ML inference and a local Qwen2.5 1.5B LLM. No cloud, no subscriptions, fully private.
Overview
What is it?
Most stock analysis tools are either cloud-dependent, subscription-gated, or black boxes. Faction is a fully local alternative: a Tauri desktop app that scores stocks on both technical (price action, candlestick patterns, momentum) and fundamental (ROE, FCF, debt/equity) dimensions using Keras models loaded directly via a pure-Rust HDF5 reader.
The AI chat feature runs Qwen2.5 1.5B locally via Hugging Face's candle inference library, Metal-accelerated on Apple Silicon, CPU elsewhere. Market data comes from Yahoo Finance with Tiingo as a fallback. Everything stays on your machine.
Features
What it does
Technical Analysis
Scores stocks based on price action (candlestick patterns, momentum indicators) using a trained Keras model running entirely on-device.
Fundamental Analysis
Evaluates financial health through key ratios (ROE, free cash flow, debt/equity) scored by a second on-device ML model.
Local AI Chat
Ask questions about any stock using Qwen2.5 1.5B running locally via candle: Metal-accelerated on Apple Silicon, CPU on everything else.
Dashboard
Combined view of technical and fundamental scores with an optional AI insight strip for a quick read on any ticker.
Fully Offline
All inference runs on your machine. No cloud API keys needed. Market data via Yahoo Finance (Tiingo as fallback), model runs locally.
Pure-Rust ML Backend
Keras .h5 models loaded directly via the hdf5 crate and ndarray: no Python runtime, no PyTorch, just native Rust inference.
See the code
Full source, setup instructions, and model details on GitHub.