hemvest AI AI Methodology
Last updated: November 2026
hemvest AI combines multiple AI models and market data sources to generate stock scores and ideas for Indian investors. This page explains the high-level methodology used to analyse securities and produce signals. It is intentionally simplified and does not reveal proprietary implementation details.
Data Sources
hemvest AI uses data from a combination of market data providers and public sources, including NSE/BSE prices, fundamentals, derivatives data, and selected third-party APIs for forecasts and sentiment. Data quality checks and sanity filters are applied before analysis.
Seven Research Layers
- Technical: indicators such as trend, momentum, support/resistance and volatility.
- Fundamental: earnings, valuations, growth, balance sheet strength and quality metrics.
- Momentum: price and volume behaviour across recent weeks and months.
- Volume: trading volume against its recent average, to spot unusual buying or selling interest.
- F&O Data: derivatives positioning, open interest and options activity where applicable.
- Sentiment: news and market tone around the stock or sector, where available.
- Delivery: delivery-based buying versus intraday churn to detect accumulation or distribution.
Score & Horizon Framework
Each stock is evaluated on the above dimensions and combined into an overall AI score between 0 and 100. The scoring process works as follows:
- Data Collection: Real-time and historical data is gathered from multiple sources including price feeds, fundamental data, derivatives markets, news sentiment, and delivery patterns.
- Dimensional Analysis: Each of the 7 research layers (Technical, Fundamental, Momentum, Volume, F&O, Sentiment, Delivery) is scored 0–100 on its own, in every horizon. Six of the seven are computed by rules from price, volume, F&O, delivery and fundamentals data; the seventh, sentiment, is read by an AI language model (Anthropic Claude). The short- and medium-term models also record a separate breakout reading, shown beside the score but not part of it.
- Weight Allocation: The layers are combined with fixed weights — in the long-term model Technical 25%, Fundamental 25%, Momentum 15%, Volume 10%, F&O 10%, Sentiment 10%, Delivery 5% — plus adjustments for analyst price targets (up to about +15 points), FII/DII flows and upcoming earnings dates.
- Score Aggregation: The weighted dimensional scores are combined using ensemble methods to produce a final AI score between 0-100.
- Horizon Mapping: The AI score is mapped to appropriate holding periods based on the underlying signals and market regime:
- 7–30 days: High momentum stocks with strong technical and short-term fundamental catalysts. These typically show immediate price action potential.
- 30–90 days: Balanced opportunities with solid fundamentals and reasonable technical setups. These allow for proper position sizing and risk management.
- 90–365 days: Deep value or growth stories with strong business fundamentals, competitive advantages, and long-term secular trends.
- Signal Generation: For each horizon, the system generates specific entry points, target prices, and stop-loss levels based on volatility, support/resistance levels, and risk-reward ratios.
The AI models are continuously retrained on new market data to adapt to changing market conditions, regime shifts, and evolving investor behavior. This dynamic approach helps maintain relevance across different market cycles.