xbtlin/ai-berkshire

What it solves

AI Berkshire provides a structured investment research framework that transforms generic AI analysis into professional-grade, decision-ready reports. It solves the problem of "balanced" but vague AI responses by forcing concrete conclusions (Pass/Fail/Grey), implementing multi-perspective adversarial analysis, and ensuring financial data precision through external Python tools rather than relying on LLM arithmetic.

How it works

The system is designed as a collection of "Skills" compatible with Claude Code and Codex. It operates across three layers:

1. Skill Layer: 20 specialized entry points for different scenarios, such as deep company research, earnings reviews, industry screening, and portfolio management.
2. Agent Layer: For complex tasks, a "Team Lead" orchestrates four distinct AI Agents representing the methodologies of value investing masters (Warren Buffett, Charlie Munger, Duan Yongping, and Li Lu). These agents independently search, analyze, and challenge each other's conclusions.
3. Tool Layer: A financial rigor toolkit (financial_rigor.py) that uses Python's decimal.Decimal for precise calculations and cross-validates data from multiple independent sources to prevent hallucinations.

Who it’s for

• Value Investors: Those following the Buffett/Munger philosophy who want to systematize their research process.
• Investment Analysts: Professionals looking to increase research depth and efficiency using AI agents.
• Retail Investors: Individuals who want a professional-grade research team experience without the overhead of a full firm.

Highlights

• Adversarial Perspectives: Simulates a debate between four investing masters to uncover blind spots.
• Anti-Bias Mechanisms: Includes information richness ratings (A/B/C), Munger-style inversion (thinking about failure), and a "Mirror Test" for decision discipline.
• Financial Rigor: Built-in tools for market cap verification, valuation checks, and Benford's Law detection to ensure data accuracy.
• Comprehensive Skill Set: Covers everything from 10-minute quick screens (/investment-checklist) to 120,000-word deep-dive series (/deep-company-series).
• Multi-Agent Parallelism: Uses parallel agents to multiply search volume and information sources compared to a single prompt.