lyra81604/zhengxi-views

What it solves

This project addresses the issue of AI hallucinations when asking about a specific fund manager's views. Instead of relying on a model's general memory, it provides a verifiable knowledge base of Zheng Xi's (E Fund Management) public investment views, ensuring that every conclusion can be traced back to a specific year, document, and original quote.

How it works

It functions as an "Agent Skill" that combines three core components:

1. Original Corpus: A collection of public texts from 2012–2026, including quarterly/annual reports, manager notes, and media interviews.
2. Distilled Methodology: A framework of investment methods derived from the corpus, allowing the AI to deduce views on new topics using Zheng Xi's logic while explicitly stating when it is a deduction rather than a direct quote.
3. Real Fund Data: Snapshots of Zheng Xi's 8 managed funds and a list of approximately 27,000 market funds, enabling real-time data fetching and comparative analysis via Python scripts.

Who it’s for

It is designed for researchers and learners who want to analyze the investment philosophy of Zheng Xi, verify his "words vs. actions" (comparing public statements to actual holdings), or use his framework to evaluate other funds.

Highlights

• Strict Traceability: Prevents fabrication by requiring citations from the original corpus for all claims.
• Methodological Deduction: Uses a distilled framework to analyze topics not explicitly mentioned in the corpus.
• Comparative Analysis: Can fetch data for any fund in the market to compare performance and style against Zheng Xi's funds.
• Multi-Platform Support: Compatible with Claude Code, Tencent WorkBuddy, Cursor, and other LLM platforms via custom instructions and knowledge files.