How Trader AI Works

Overview

Trader AI uses a three-stage analysis pipeline combined with six intelligence data sources to find cheap stocks that are trending upward. Every day after market close, the system scans all tracked stocks and ranks them by opportunity score.

The system runs locally on a private server using Llama 3.2 AI for reasoning, XGBoost machine learning for price prediction, and pandas-ta for technical indicators. No cloud APIs, no subscription fees — all analysis happens on our own hardware.

Stage 1: Technical Indicators

Computed for every stock using 2 years of historical price data:

RSI (Relative Strength Index)
Measures if a stock is overbought (>70) or oversold (<30). Oversold stocks near their bottom are potential buys.

MACD (Moving Average Convergence Divergence)
Detects momentum shifts. A bullish crossover (MACD crossing above signal line) indicates upward momentum is building.

SMA 20 / 50 / 200 (Simple Moving Averages)
Shows short, medium, and long-term trends. Price below SMA200 but above SMA20 = long-term cheap but recovering.

Bollinger Bands
Measures volatility and price extremes. Stocks near the lower band are potentially oversold.

ATR (Average True Range)
Measures daily price volatility. Used to set stop-loss and target price levels.

Volume Analysis
Compares current volume to 20-day average. High volume on up days signals institutional buying (accumulation).

Stochastic Oscillator
Another momentum indicator that confirms overbought/oversold conditions alongside RSI.

52-Week Position
Where the current price sits relative to its yearly high and low. Closer to the low = cheaper = higher opportunity.

All indicators are combined into a Technical Score ranging from -100 (extremely bearish) to +100 (extremely bullish).

Stage 2: Machine Learning Prediction

An XGBoost gradient boosted model is trained individually for each stock on its own price history. The model learns stock-specific patterns from:

1-day, 5-day, and 20-day price returns
5-day and 20-day realized volatility
Volume ratio (current vs. average)
RSI, MACD histogram, Bollinger %B values
Gap analysis (open vs. previous close)
Day-of-week cyclical patterns

Output: Predicted 5-day price direction (UP or DOWN) with a confidence percentage. Models are retrained weekly on the latest data.

Stage 3: AI Reasoning (Llama 3.2)

The top 10 candidates from stages 1 and 2 are sent to a Llama 3.2 3B large language model running locally on GPU. The AI receives ALL the data — technical indicators, ML prediction, and all intelligence signals — and provides:

Signal: STRONG BUY, BUY, HOLD, or SELL
Confidence: 0-100% based on data quality and signal alignment
Target Price: 4-week price target
Stop Loss: Recommended exit if the trade goes wrong
Recommended Amount: How much to invest ($500-$2,500 from a $10,000 daily budget)
Reasoning: 2-3 sentence explanation referencing the specific data points

6 Intelligence Data Sources

Beyond price data, we collect free intelligence that most retail traders don't use:

1. Insider Trades (SEC EDGAR)
When a CEO, CFO, or board member buys their own company's stock with their own money, it's the strongest bullish signal in finance. They know the company better than anyone. We track all Form 4 filings from the SEC.

2. Congressional Trades (Public Record)
US politicians are required to disclose their stock trades. They often trade ahead of legislation and regulatory changes. We monitor House and Senate stock transaction disclosures.

3. News Sentiment Analysis
We scrape financial news headlines from Yahoo Finance RSS feeds for every tracked stock and analyze whether the sentiment is positive, negative, or neutral. Positive news + cheap price = opportunity.

4. Short Interest (Yahoo Finance)
Stocks with high short interest (many investors betting against them) combined with upward momentum can trigger a short squeeze — explosive upward price moves as short sellers are forced to buy back shares.

5. Earnings Surprises
Companies that consistently beat earnings estimates tend to continue outperforming. We track quarterly earnings beats/misses. A company that beats estimates 3+ quarters in a row trading at a discount = quality on sale.

6. Market Sentiment (VIX / Fear & Greed)
The VIX "fear index" measures overall market anxiety. When VIX is high (fear) and an individual stock is cheap but recovering, that's historically the best time to buy. We boost opportunity scores during fearful markets.

Opportunity Score (0-100)

Every stock gets a composite score. Higher = better buy opportunity:

52-Week Position	20%	How close to yearly low (cheaper = higher score)
Insider Buying	15%	Recent insider purchases (strongest signal)
SMA Recovery Pattern	15%	Below long-term average but recovering short-term
RSI Oversold	10%	RSI below 30-40 = potentially oversold
MACD Momentum	10%	Bullish crossover or approaching crossover
News Sentiment	10%	Positive headlines vs. negative
Earnings Beats	5%	Consistent earnings surprises
Congress Buying	5%	Politicians purchasing the stock
Short Squeeze Potential	5%	High short interest + upward momentum
ML Prediction	5%	XGBoost model confidence in upward move

During fearful markets (high VIX), all scores get a 10% bonus — fear creates the best buying opportunities.

Position Sizing

The AI recommends how much to invest per pick from a $1,000 daily budget based on signal strength and confidence:

STRONG BUY 80%+	$250	Highest conviction — multiple strong signals aligned
STRONG BUY 60-79%	$200	Strong but some uncertainty
BUY 80%+	$150	High confidence buy
BUY 60-79%	$100	Standard position
BUY <60%	$50	Small speculative position
HOLD / SELL	$0	No investment recommended

Daily Schedule (All Times ET)

9:30 AM - 4:00 PM	Live quote refresh every 5 minutes
4:30 PM	End-of-day price data collection for all stocks
4:45 PM	Intelligence data collection (insider trades, congress, news, earnings, sentiment)
5:00 PM	Daily AI Market Scan — screens all stocks and generates top 10 recommendations
Saturday 2 AM	Weekly ML model retraining on latest data

Technology Stack

AI Model: Llama 3.2 3B (local GPU)

ML Engine: XGBoost + scikit-learn

Technical Analysis: pandas-ta

Backend: Python FastAPI

Database: MySQL 8.0

Stock Data: Yahoo Finance (free)

AI Runtime: Ollama

Android App: Kotlin + Jetpack Compose

GPU: NVIDIA Quadro P2000