SENTRAL - Multi Spectrum Stock Analysis

Links: GitHub repository

SENTRAL is an end-to-end, open-source stock analysis platform that blends classical financial analysis with modern LLM-driven sentiment and machine-learning forecasting to produce a single, explainable buy, hold or sell signal. It ships as two production-ready Streamlit apps (SENTRAL for deep single-stock analysis and a companion SCREENER inspired by screener.in and groww.in), backed by an 82-cell analysis notebook.

The analysis pipeline

• Fundamental scoring: discounted cash flow (DCF), the Altman Z-Score, the Piotroski F-Score, Graham valuation and a wide set of financial ratios computed from statements.
• Technical analysis: 35 indicators (moving averages, RSI, MACD, ATR and more) alongside risk metrics, chart-pattern detection and seasonality analysis.
• News and sentiment: news is gathered from 13 sources (APIs, RSS feeds and Reddit), filtered for relevance, then scored by 10 sentiment models, including finance-tuned and reasoning LLMs.
• ML forecasting: LSTM and Transformer models forecast price trends, complemented by Monte Carlo simulation for uncertainty.
• Signals and backtesting: a composite engine fuses fundamentals, technicals and sentiment into a BUY/HOLD/SELL signal, and a 20-strategy backtesting module validates the approach before any decision.
• Reporting: every run produces an HTML and PDF report plus the underlying artefacts (forecasts, backtests, news corpus and trained models).

Validation before the signal

No recommendation ships unbacktested. Twenty strategies are replayed over the price history under explicit assumptions (0.1% commission round-trip, fully invested, close-price execution), and each reports Total Return, CAGR, Sharpe, Sortino, maximum drawdown, win rate, profit factor, trade count, Kelly criterion and a suggested stop-loss. Correlation and feature-importance visualisations sit behind each recommendation, so the composite signal can be traced back to the pillars that moved it.