Yogesh-VG0/Stock_Predict_Ai
AI-powered stock market prediction platform with real-time analytics, news, and advanced visualization tools.
What's novel
AI-powered stock market prediction platform with real-time analytics, news, and advanced visualization tools.
Code Analysis
8 files read · 3 roundsAn AI-powered stock prediction platform that uses LightGBM with LSTM temporal embeddings to predict 1/7/30-day market-neutral alpha for S&P 75 companies, with cross-sectional ranking, SHAP explanations, and a full Next.js/Node.js/MongoDB stack serving pre-computed predictions.
Strengths
Exceptionally deep ML engineering: leakage-proof feature engineering with 77+ features, walk-forward validation with purge/embargo, prediction shrinkage, confidence calibration, per-ticker quality gates, and a well-designed LSTM-as-feature-extractor architecture. The codebase shows genuine iterative refinement (v7→v10.3) with detailed comments explaining every parameter change and its rationale.
Weaknesses
Test coverage is minimal (only one security test file visible in the backend), and several core modules are extremely large single files (predictor.py at 114KB, sentiment.py at 143KB, api/main.py at 151KB) which hurts maintainability. The cross-sectional ranking and shrinkage logic, while well-intentioned, add significant complexity that makes the prediction pipeline hard to debug or extend.
Score Breakdown
Signal breakdown
Innovation
Craft
Traction
Scope
Evidence
Commits
241
Contributors
1
Files
171
Active weeks
7
Repository
Language
Python
Stars
3
Forks
1
License
AGPL-3.0