Institutional Documentation & Support

Explore mathematical specifications, causality test documentation, and system architecture.

5-Model NLP Ensemble

Multi-layered neural network sentiment extraction combining financial FinBERT, contextual RoBERTa, and generative FinGPT.

Hybrid Price Forecaster

Dual-engine quantitative architecture pairing macro structural Prophet curves with recurrent LSTM neural networks.

7-Factor Fear & Greed

Real-time composite sentiment gauging volatility regimes, put/call ratios, stock price breadth, and safe haven flows.

Frequently Asked Questions (FAQ)

How does the 5-Model NLP Sentiment Ensemble operate?

Indra-MarketMind processes news headlines and social feeds through FinBERT (35%), RoBERTa (25%), FinGPT (20%), VADER (10%), and TextBlob (10%). The scores are weighted and calibrated against historical volatility to generate normalized sentiment scores between -1.0 (Extreme Bearish) and +1.0 (Extreme Bullish).

What methodology is used for price forecasting?

We combine Facebook Prophet for macroeconomic seasonality/regime shifts with a bidirectional deep LSTM network for short-term momentum. The output produces confidence intervals with Monte Carlo fan cones for 7 to 90 days.

Where is market data sourced from?

Real-time quotes, sector indexes, and benchmark feeds are ingested via multi-provider failover bridges including Yahoo Finance, Finnhub, NSE/BSE institutional webhooks, and SEC EDGAR filings.

Are my custom API keys stored securely?

Yes. All custom API keys entered in the Settings Vault are isolated in your browser's local sandbox storage using AES-256 client isolation and are never stored on unauthenticated servers.