

Make Your Own Financial Agent for Investors
Agenda Overview
Section 1: Agent Architectures:
- Understand basic agentic AI concepts: single vs. multi-agent systems.
- Explore ReAct (Reason-Act-Observe) loops for decision-making and function APIs for data access.
- Demonstration: Use a LangGraph agent to query Yahoo Finance for cap table stress-tests.
- Assignment: Customize a GitHub template for your investment sectors.
- Demystify AI agents — break down LLMs, memory, tools, and the Observe → Think → Act loop.
- Compare architectures — monolithic vs. multi-agent, and heavyweight frameworks vs. minimalist harnesses.
- Explore pi-mono — a "primitives, not features" open-source toolkit for building custom agents.
- Build live — code a real VC due diligence research agent from scratch in the final session.
Section 2: Automated Deal Sourcing:
- Create agents for lead generation using vector embeddings on Crunchbase/LinkedIn data.
- Discuss prompt engineering to evaluate founder teams.
- Hands-on integration: Combine Airtable with OpenAI agents for real-time matching; assess performance using precision and recall metrics.
Section 3: Advanced Due Diligence
- Implement hierarchical agents for financial analysis and risk management.
- Focus on memory storage for audit trails and human oversight.
- Demo: Backtest your portfolio for performance improvements; homework involves a diligence simulation on live data.
**Section 4: Portfolio Agents & Scaling (2.5 hours)**
- Coordinate multi-agent systems for monitoring, exit signals, and rebalancing.
- Discuss costs, addressing errors, and deployment methods.
- Capstone project: Collaboratively build a "VC Vault Agent"; participants receive code repositories, certificates, and personalized tuning sessions