What is it?

This repository is a sophisticated "Agent Skill" designed to transform any LLM (like Claude, GPT, or Gemini) into a research assistant that understands the specific investment philosophy of Zheng Xi (a prominent fund manager at E Fund). Unlike standard RAG (Retrieval-Augmented Generation) implementations that might simply summarize text, this tool focuses on traceability.

It integrates three pillars of data: a comprehensive corpus of original public views (2012–2026), a distilled investment methodology framework backed by verbatim evidence, and real-time fund data covering approximately 27,000 funds across the market. The result is an AI that doesn't just tell you what a manager "might" think, but tells you exactly what they said, when they said it, and whether their actual portfolio holdings prove it.

Key Features & Use Cases

The system is designed for high-precision financial analysis where "hallucinations" (AI-generated falsehoods) are unacceptable. Here are the concrete ways developers and analysts can utilize it:

• Evidence-Based Q&A: Instead of general summaries, the skill retrieves specific quotes from quarterly reports, manager notes, and media interviews to answer questions about sector outlooks.
• Methodological Deduction: If a manager has never mentioned a specific sector (e.g., innovative drugs), the system uses a "distilled framework" to simulate how that manager would analyze the sector based on their known logic, while explicitly stating that the output is a deduction, not a direct quote.
• Consistency Auditing (Say-Do Gap): Users can cross-reference public statements with actual portfolio holdings. For example, if a manager claims to be bullish on optical communications, the tool checks if the top holdings actually reflect that sentiment.
• Comparative Scoring: Using a proprietary six-dimensional scoring card, the tool can analyze any fund in the market to determine how closely its style aligns with the manager's specific investment criteria.
• Style Mimicry: It can generate market commentaries that mimic the professional tone and structure of specific fund reports without fabricating financial figures.

Why is it trending?

The project addresses a critical pain point in the AI-driven financial sector: the reliability of information. Most LLMs struggle with the "last mile" of financial accuracy, often blending different managers' views or inventing quotes. By providing a structured "Skill" that separates original corpus from distilled logic and hard data, it creates a gold standard for how investment research agents should operate.

The community is reacting positively to its "zero-fabrication" approach. The growth in stars reflects a growing demand for "Vertical AI"—tools that don't try to know everything, but know one specific domain with absolute precision and full traceability.

Who should use it?

This tool is ideal for a variety of users within the financial and tech ecosystems:

• Equity Researchers: Who need to track the evolution of a manager's views over a decade without manually reading hundreds of PDFs.
• Retail Investors: Who want to understand the "logic" behind a fund's performance rather than just looking at a returns chart.
• AI Developers: Who are building financial agents and want a blueprint for how to implement a "traceable" knowledge base that prevents hallucinations.
• Quantitative Analysts: Looking for a way to blend qualitative manager views with quantitative fund data.

Since it is written in Python and designed as an Agent Skill, it fits perfectly into modern AI workflows involving Cursor, ChatGPT, or specialized AI assistants like ima and WorkBuddy.

Getting started

To integrate this skill into your AI workflow, follow these basic steps:

1. Ensure you have a Python environment installed.
2. Clone the repository to access the references/ directory containing the corpus and fund data.
3. Import the references/corpus/ for original texts and references/method.md for the logic framework.
4. Configure your AI platform (Claude, ChatGPT, etc.) to use these files as the primary knowledge base.
5. Query the agent using specific prompts such as: "Using Zheng Xi's framework, score the [Fund Name] and explain the reasoning."

For the full setup guide, see the official repository.