Institutional fundamental research requires analysts to gather evidence, identify business drivers, compare competing viewpoints, and generate investment memos. According to a recent arXiv paper by Di Zhu, Lei Nico Zheng, and Zihan Chen, this process is qualitatively distinct from trading-signal generation and is better served by an independence-preserving architecture. The authors present FundaPod, a multi-persona agent platform for AI-assisted fundamental investment research.
FundaPod features AI agents with different personas—such as value investors or macro strategists—that conduct research independently under a shared provenance contract. Their disagreements are surfaced post hoc for adjudication by the human portfolio manager (PM) through a knowledge-graph memory system. This system connects tickers, memos, analysts, and themes, acting as a "second brain" for cumulative investment knowledge.
The paper contributes five design principles for human-AI hybrid systems supporting fundamental research, grounded in design-science practice and theories of cognitive isolation and human-machine coordination. It also describes four architectural mechanisms:
| Mechanism | Description |
|---|---|
| Persona distillation pipeline | Turns public investor materials into deployable agents |
| Declarative skill registry | Lets the planner derive typed task graphs |
| Grounded evidence model | Links memo claims to verifiable sources |
| Knowledge-graph "second brain" | Connects tickers, memos, analysts, and themes |
The platform is designed to produce investment plans that are transparent, reusable, and verifiable, contributing to the cumulative development of investment knowledge. The authors argue that fundamental research is a human-centric decision-support task, and FundaPod's architecture preserves independence among agents while enabling oversight.
The paper includes a complete case study and a persona-based memo comparison to demonstrate the architecture. The authors emphasize that FundaPod's goal is not merely to predict outcomes but to produce testable investment plans.
For enterprise technology decision-makers, FundaPod illustrates how multi-agent AI systems with knowledge graph memory can support complex, human-centric analytical tasks. The design principles—grounded in cognitive isolation and human-machine coordination—offer a framework for building AI assistants that augment rather than replace expert judgment. The architectural mechanisms, such as the persona distillation pipeline and grounded evidence model, provide concrete patterns for integrating LLMs into institutional workflows where transparency and verifiability are critical.