xbtlin/ai-berkshire

What it solves

AI Berkshire addresses the lack of decision-grade quality in standard AI financial analysis. While general LLMs often provide balanced but vague summaries, this framework enforces decision discipline, multi-perspective adversarial analysis, and financial rigor to produce actionable investment research reports rather than generic summaries.

How it works

The project provides a collection of structured "Skills" (commands) compatible with Claude Code and Codex. It transforms the AI from a single analyst into a virtual investment team by:

• Multi-Agent Adversarial Perspective: It simulates four value investing masters (Warren Buffett, Charlie Munger, Duan Yongping, and Li Lu), forcing them to challenge each other's conclusions (e.g., one focuses on business model, another on risk/failure scenarios).
• Structured Workflows: It uses specific pipelines for different tasks, such as the "Industry Funnel" (filtering the whole market down to 3 companies) or the "Investment Checklist" (a 6-step binary filter).
• Financial Rigor Tools: It uses a dedicated Python tool (financial_rigor.py) employing decimal.Decimal for precise calculations to avoid LLM arithmetic errors in market cap and valuation.
• Anti-Bias Mechanisms: It implements information richness ratings (A/B/C) and "Mirror Tests" to ensure the researcher can articulate the investment thesis clearly or admit ignorance.

Who it’s for

Value investors and financial analysts who want to leverage AI to automate the deep-dive research process while maintaining the strict intellectual discipline of professional value investing.

Highlights

• 20 Specialized Skills: Includes deep research, earnings reviews, industry scanning, portfolio management, and news attribution.
• Multi-Agent Parallelism: The /investment-team skill deploys four independent agents to research a company simultaneously, providing four times the search volume and diverse perspectives.
• Decision-First Output: Forces the AI to provide concrete conclusions (Pass/Fail/Grey) and specific price ranges rather than neutral summaries.
• Private Company Framework: A specialized "detective-style" research flow for companies with scarce public data.
• News Pulse: A rapid 10-15 minute attribution tool to distinguish between fundamental value events and emotional market noise.