Skip to main content
Preisser Solutions
Case Studies/Internal Platform
Internal Platform • AI Architecture

Alpha Matrix — A Self-Evolving 6-Agent Stock Analysis Platform

Six specialized agents run a daily pipeline against the entire US equity universe. Each agent maintains a soul file that evolves based on prediction accuracy.

6
Specialized agents in the pipeline
Daily
Full pipeline run cadence
Soul files
Per-agent evolution based on accuracy
US equity
Universe scanned each morning
01
6
Specialized agents in the pipeline
02
Daily
Full pipeline run cadence
03
Soul files
Per-agent evolution based on accuracy
04
US equity
Universe scanned each morning
Before

A daily research workflow that wanted to be a pipeline.

Tyler built Alpha Matrix as an internal platform — a daily pipeline against the US equity universe with one rule: every agent has to learn from its own track record. Pre-existing tools generate analysis, but they do not remember what worked and what didn't. They produce confident output every morning regardless of whether yesterday's confident output was right.

Alpha Matrix is built around a different assumption. Every prediction gets logged. Every outcome gets resolved. Every agent's soul file is updated based on what its calls actually did. The system gets better at noticing what it is bad at — and gets honest about it.

What we built

Six specialized agents, one daily pipeline, persistent soul files.

Tyler built this internal trading platform as a six-agent pipeline. A Scanner walks the entire US equity universe pre-market and flags anomalous signals. A Sentiment Engine analyzes Reddit, news, analyst data, and social perception. A Quant Analyzer ingests price, fundamentals, and options data. A Pattern Matcher correlates historical analogs with the scanner, sentiment, and quant outputs. A Conviction Scorer synthesizes everything into confidence ratings with trade theses. A Portfolio Strategist applies position sizing, risk management, and a daily top-five output.

Data flows from Polygon.io, Alpha Vantage, SEC EDGAR, FRED, Reddit, NewsAPI, Finviz, Unusual Whales, Google Trends, and Stocktwits. Every prediction the system makes is logged. Outcomes are resolved against the prediction log. Each agent's soul file gets rewritten based on what worked — winners get reinforced, losers get demoted. The architecture pattern is what makes the platform worth talking about: the same self-evolving multi-agent design re-applies to any analytical domain that produces predictions and resolves outcomes.

Specifications

Pipeline architecture.

Six specialized agents run sequentially in a daily pipeline

Each agent maintains a soul file that evolves based on prediction accuracy

Predictions are logged; outcomes are resolved against the log

Conviction matrix weights are adjusted run-over-run based on signal performance

Output: daily top-five categories with trade theses and risk framing

The six agents

  • Scanner — pre-market scan of the US equity universe, flags anomalous signals
  • Sentiment Engine — Reddit, news, analyst data, social perception
  • Quant Analyzer — price, fundamentals, options data
  • Pattern Matcher — historical analogs, correlates scanner + sentiment + quant
  • Conviction Scorer — synthesizes into confidence rating with trade theses
  • Portfolio Strategist — position sizing, risk, daily top-five categories

Data sources

  • Polygon.io — market data
  • Alpha Vantage — equity fundamentals and intraday
  • SEC EDGAR — filings
  • FRED — macro economic indicators
  • Reddit API and NewsAPI — sentiment inputs
  • Finviz, Unusual Whales — options and screener data
  • Google Trends and Stocktwits — behavioral signals
Results

Outcomes the engagement actually produced.

Result 01
6
Specialized agents in the pipeline

Scanner, Sentiment, Quant, Pattern, Conviction, and Strategist run in sequence every day against the US equity universe.

Result 02
Daily
Full pipeline cadence

The full pipeline runs on a daily cadence — pre-market scan through post-market archival and learning.

Result 03
Soul files
Per-agent evolution

Each agent maintains a persistent soul file rewritten based on prediction accuracy — winners reinforced, losers demoted.

Result 04
10+ sources
Data sources orchestrated

Polygon.io, Alpha Vantage, SEC EDGAR, FRED, Reddit, NewsAPI, Finviz, Unusual Whales, Google Trends, and Stocktwits feed the pipeline.

Tech stack
Claude APIMulti-agent orchestrationPolygon.ioAlpha VantageSEC EDGARFREDReddit APINewsAPIFinvizUnusual WhalesGoogle TrendsStocktwits

Need a self-evolving multi-agent pipeline for your domain?

Preisser Solutions builds the same multi-agent architecture pattern for any analytical domain that produces predictions and resolves outcomes. Free 30-minute scoping call.