How AI-Stock picks stocks
A plain-language explanation of the XGBoost model, FinBERT sentiment layer, and scoring system that generates every daily pick.
Three signal layers
1. Technical analysis
For every NSE stock the model computes RSI (momentum and overbought/oversold conditions), ATR (stop sizing relative to volatility), moving-average crossovers, volume-weighted averages, and proprietary momentum scores from recent price action. Stocks with average daily volume below ₹2 Cr are filtered out before scoring.
2. Fundamental screening
Technical signals are overlaid with quarterly filing data: revenue growth (YoY and QoQ), debt-to-equity ratios, return on equity, and operating margin trends. Strong technical momentum with deteriorating fundamentals receives a composite score penalty, reducing its chance of appearing as a top pick.
3. FinBERT news sentiment
Each morning, recent Google News headlines are fetched for every stock in scope and passed through FinBERT — a BERT-based transformer fine-tuned on financial text. FinBERT assigns a sentiment score (positive, neutral, or negative) to each headline. The aggregate sentiment is folded into the composite score as a third signal layer.
The XGBoost model
XGBoost (Extreme Gradient Boosting) is an ensemble machine learning algorithm that combines many small decision trees into a single strong predictor. AI-Stock trains this model on historical NSE data where the target label is whether a stock outperformed the Nifty 50 over the following 5 trading days.
Each stock receives a composite score from 0 to 100. A score above 65 is required to be classified as STRONG BUY and eligible as a daily pick. The three highest-scoring eligible stocks are selected.
Confidence tiers
All three signal layers — technical, fundamental, and sentiment — agree on a bullish signal. Highest-confidence picks.
Technical signal is strong and at least one of fundamental or sentiment supports the thesis. Moderate confidence.
Signal driven by technical indicators alone. Fundamental or sentiment data unavailable or neutral. Consider smaller position sizes.
How picks are validated
Every pick is evaluated exactly 5 trading days after publication. The exit price is fetched automatically and compared against the Nifty 50 return over the same period. A pick is marked as a hit if the stock outperformed Nifty by at least 1% (after accounting for transaction costs). The result — hit or miss — is published on the public track record page.
See the full track recordFrequently asked questions
How does AI-Stock pick stocks?
An XGBoost model trained on historical NSE data combines technical indicators, fundamental data, and FinBERT news sentiment into a composite score. The top 3 stocks scoring above 65 are selected each trading day.
What is FinBERT and why use it?
FinBERT is a BERT-based transformer fine-tuned on financial text. It reads news headlines for each stock and assigns a sentiment score — adding a news signal layer that pure technical models miss.
How accurate are the picks?
Every pick is evaluated 5 trading days after publication and the result is published publicly. You can see the full unfiltered track record — every pick ever made, every outcome — on the Track Record page.
Is AI-Stock SEBI-registered?
No. AI-Stock is not a SEBI-registered research analyst or investment advisor. The picks are informational research signals only and do not constitute investment advice. Always consult a registered advisor before investing.
Risk disclaimer
AI-Stock is not registered with SEBI as a research analyst or investment advisor. Signals are for informational purposes only and do not constitute investment advice. Investing involves market risk including possible loss of principal. Past accuracy does not guarantee future performance.