Watchlist-only until validated
NSE Quant Engine
A local, offline-first screener and analytics engine for NSE equities and ETFs, with a PySide6 desktop interface. This page is only the project landing page — the engine itself runs on your machine.
Honest limitation. This is a screener that, once validated, can tell you whether its ranking beats a benchmark after costs. It is not a low-risk, high-profit oracle. Its most valuable possible answer is often “No Proven Edge Yet.” Research and educational use only — not investment advice.
Architecture in one line
Python computes. Validation measures. AI explains. The human decides.
Python computes
Every score, rank, factor and cost estimate comes from deterministic Python. No model guesses a number.
Validation measures
One question: does the ranking beat the benchmark after costs, out of sample? Walk-forward, shrunk, cost-aware.
AI explains
An optional evidence bundle lets an LLM narrate what the numbers already say. Explanation never feeds back into scores.
The human decides
No broker integration, no orders. A screener with an audit trail — and a loud watchlist-only state.
Validation philosophy
- One source of truth.
validation_status.jsonholds the verdict. Reports are never scraped for it. - One ranking authority. Confidence-adjusted score descending, symbol ascending. Raw score rank is diagnostic only.
- Evidence, not vibes. Walk-forward evaluation, an incremental residual-IC gate for new alphas, Bayesian shrinkage on IC and hit rate, costs included.
- Context stays context. News and filings are shown for human review and never touch scores or ranks.
- Dormant by default. Adaptive weighting ships disabled and must be enabled deliberately.
Desktop application
Install and run
cd desktop/nse_quant_engine
python -m venv .venv
pip install -r requirements.txt
python run_app.py
# headless full pipeline
python orchestrator.py --allRequires Python 3.11 or 3.12. Runtime caches and results are written locally and never published.
What a run produces
- validation_status.json
- Canonical, authoritative verdict
- latest_scores.xlsx
- Official scores and ranks
- latest_scores_v4_shadow.xlsx
- Shadow scores (never reorders official)
- daily_changes.json
- Entrants, exits, rank gainers, new risk flags
- news_digest.json
- Context-only news and exchange filings
- dashboard_latest.html
- Dashboard, embedded in the desktop app
Synthetic examples of these shapes live in examples/sample_output/. No real market data, validation history or portfolio file is published.