The flagship program
Mastering AI Agents in Finance
One path, two steps: a self-paced foundation you can start today, then the flagship cohort, Architecture to Production, that takes you from multi-agent design to production systems a risk committee can audit. Everything is built the same way I learned this myself: raw code first, so you understand what the frameworks are doing before you let them do it for you.
The path
Start self-paced. Finish in the flagship cohort.
Two steps. Level 1 gets you building agents from scratch, on your own schedule: it is the on-ramp and the prerequisite. The flagship cohort takes you the rest of the way: from multi-agent architectures you can defend in a design review to systems that are guarded, traced, and evaluated, the kind your institution will actually run.
AI Agents in Finance: The Complete Foundations
From your first LLM API call to your first multi-agent finance system. For Python-capable finance and tech professionals starting from zero on LLMs: you hand-build your own agent harness, the loop that turns a language model into something that acts, before touching any SDK, so when the framework arrives, nothing in it is magic.
- Hand-build tool calling, the agent loop and ReAct, then rebuild on the OpenAI Agents SDK
- Memory that persists across runs + agentic RAG with LlamaIndex
- Multi-agent handoffs, routing, and agents-as-tools
- Deep dives: Morning Briefing Desk (Espresso) · Financial News Bot (Hot Take) · Earnings Season Watchtower (Harvest)
AI Agents in Finance: Architecture to Production
You've built an agent. This is the rest of the road, in one program. First the architecture: systems of agents you can defend in a design review, with stateful graphs that branch, persist, and resume; reasoning patterns that check their own work; memory and context engineering; orchestration with human approval gates; and agentic RAG built two ways. Then the production discipline no demo survives without: harness and loop engineering, guardrails, security and auditability, tracing at scale, and evaluation in depth.
- Think in graphs, not loops: typed state, branching, checkpointing, mid-run resume
- Multi-agent orchestration: evaluator-optimizer, handoffs, agents-as-tools, approval gates
- Guardrails, injection defense, audit trails, and end-to-end observability
- Evaluation in depth: LLM-as-a-judge, tool-call and trajectory evals, RAG evaluation
Self-assessment
Where do I start?
Thirty seconds of honesty saves you a wrong purchase. Join where the description matches you.
You've never built an agent
You write Python, you follow the AI news, but you've never wired a model to a tool and watched it act. Start at Level 1: Foundations. You'll build the agent loop by hand before any framework touches your keyboard.
You've built one; now you want systems you can defend and ship
You've shipped a single agent or a chatbot with tools, but multi-agent design, memory, and production hardening are still improvisation. Join the flagship cohort: Architecture to Production. You'll learn the patterns, when each one earns its complexity, and the guardrails, tracing, and evaluation that turn a demo into something a risk committee can audit.
It's not just you: your team needs this
If you're bringing a whole team or a leadership committee rather than yourself, the right door is corporate training: one in-company program, a technical track and a leadership track, delivered on-site or remote.
How I teach
The pedagogy, stated plainly
These four choices shape every chapter, from the first self-paced lab to the cohort's capstone. If you disagree with them, this program isn't for you; better to know now.
Raw first, then SDK
You build tool calling, the agent loop, and RAG in plain Python before any framework. When abstractions break in production, and they do, you'll know exactly what's underneath.
Finance-native labs
No pizza-order tutorials. Every lab is a desk you could sit at: briefing desks, credit committees, FX strategy, robo-advisory. The domain friction is the point.
Build-up, not toy-hop
Systems grow chapter over chapter instead of resetting to a new toy every lesson. You finish each level with something coherent, not a folder of disconnected demos.
Notebooks + script mirrors
Every lab ships as a Colab-ready notebook and a mirrored Python script: explore in the notebook, then see the same code structured the way production code actually looks.
Questions
Frequently asked
What are the prerequisites?
Working Python: you can write functions, use dictionaries, install packages, and read a stack trace without panic. That's it for Level 1; no prior agent or ML experience is assumed. The flagship cohort assumes you've built at least one agent (from Level 1 or elsewhere). No finance background is required, though the labs will feel more familiar if you have one.
Do I need Level 1 before the cohort?
Level 1 is the on-ramp, not a toll gate. If you've already built agents on your own, join the cohort directly; if you've never wired a model to a tool, take Level 1 first. The self-assessment above is an honest guide to where you belong.
What is the refund policy?
Level 1 (self-paced) carries a 30-day money-back guarantee. The flagship cohort offers a full refund through the end of week 1: enough time to attend the first live call, work the first labs, and decide with real information. No forms to justify, no friction.
When does the next cohort start?
Level 1 is self-paced; start whenever you enroll. The flagship cohort runs once per season, and dates are announced to the waitlist first. Join the waitlist to get the next dates and the first-15 enrollment bonuses.
Will this teach me to trade or pick stocks?
No. Let me be explicit: this program teaches you to engineer AI agent systems for financial workflows: research, analysis, document intelligence, advisory tooling, monitoring. It does not teach trading strategies, stock picking, or any way to make money in markets, and nothing in it is investment advice. If that's what you're looking for, this is not your course.
Ready?
Two steps. Start today.
From your first hand-built agent loop to a deployed, guardrailed system; the path is laid out. The only question is where you enter it.
Not sure? Take the 30-second self-assessment or ask me directly.