Skip to content

About

No description, website, or topics provided.

Resources

Stars

0 stars

Watchers

0 watching

Forks

 
 

Repository files navigation

AI FDE: The Forward Deployed Engineer & Agentic AI Architect

UnderCover Agent: the Travel Insurance Platform

An alien in disguise joins a line of travelers

Welcome to the most substantive commercial project I've ever made on any course! We will work on this project throughout the 3-week journey, culminating in a capstone that is truly enterprise grade.

This is the repo that accompanies the course. UnderCover is a fictional travel insurer. Over the three weeks we build its claims platform and an AI agent that handles claims, and in Week 3 we deploy it all to AWS.

If you're looking at this in VS Code or Cursor, right-click README.md in the Explorer on the left and choose Open Preview to see it formatted.

Key links

And in case you haven't heard it from me enough, I'd be delighted if you'd consider connecting or following on LinkedIn or subscribing on YouTube.

A note on API costs

While about $10 of API spend would be typical for this course, there are free alternatives for everything. Most labs have a free version whose name ends in free, the installation steps include a free coding agent, and in Week 3 the free route uses Render in place of AWS.

For immediate help

Just ask your coding agent (Cursor, Claude Code, etc.). I've equipped it with full knowledge of the course, the labs and this project, and it can help!

Sharing your work

I'd love to see what you build! Please share it with a pull request to community_contributions. The steps for that, and for pulling the latest code, are in guides/for_agents/AGENT_GIT_GUIDE.md, and your coding agent can take you through them.

The journey

Week 1 Day 1 is in the setup/ folder and Week 1 Day 2 is in labs/. From Week 1 Day 3 onwards we work in the entire repo. The installation steps for each day are in setup/INSTRUCTIONS.md, and the guides folder has guides to the command line, git, Python and other foundations if you'd like a refresher.

Switching to my version

You can follow the whole course on main with your own build, without ever switching branches. On each day that we build the product with a coding agent, my version is on two branches:

  • week1day5-start and so on: my build so far, with the day's product brief in docs/
  • week1day5-end and so on: my build at the end of the day

Switch to one with git checkout, for example git checkout week1day5-end. If you carry on with your own build instead, copy the day's brief from labs/ into docs/ yourself, for example labs/week1day5_product_HARNESS.md to docs/HARNESS.md.

Commit your own changes before you switch branches. Your .env file isn't in git, so it stays as it is when you switch.

Week 1: Become an AI Engineer

Day 1: Path to FDE

Day 2: Frontier LLMs & Agents

Day 3: Coding Agents & ADLC

  • Installing a coding agent: setup/INSTRUCTIONS.md
  • Your coding agent builds the pilot on main from the documents in docs/
  • My build: git checkout week1day3-end

Day 4: Open-Source, Local, Token Cost

Day 5: Harness Engineering

Week 2: Become an AI Architect

Day 1: Agent Architecture, Evals, Observability

Day 2: Agentic RAG, MCP, Memory

Day 3: Durable State & Human in the Loop

Day 4: Agent Swarms

Day 5: Self-Improvement, RSI

Week 3: Become an AI FDE

This week UnderCover goes into production on AWS, and the free route uses Render. To see what you're spending on AWS, run uv run monitor.py.

Day 1: AI Solutions Architect

Day 2: Security, Guardrails, API Gateways

Day 3: Enterprise Context Layer & Integrations

Day 4: Production Deployment

Day 5: Capstone

The branches

Command What it gets you
git checkout main The starting point: setup, labs, guides, data, and the product's documents and skeleton before the Week 1 Day 3 build
git checkout week1day3-end The pilot: a FastAPI backend, a dashboard for claims handlers and a first agent
git checkout week1day5-start My build so far, with the brief in docs/HARNESS.md
git checkout week1day5-end The agent's harness: its tools, middleware and instructions, and the claims manual as documents it can search
git checkout week2day3-start My build so far, with the brief in docs/DURABLE.md
git checkout week2day3-end Durable state: Supabase, the claim agent as a LangGraph graph that carries on after a restart, and the payout rules
git checkout week2day4-start My build so far, with the brief in docs/SWARM.md
git checkout week2day4-end The swarm: a dispatcher, started from the dashboard, that works the claims with up to 25 claim agents in parallel
git checkout week2day5-start My build so far, with the brief in docs/IMPROVE.md
git checkout week2day5-end Self-improvement: the agent's configuration in versions, an improver agent whose changes a handler approves, and sign-in for handlers
git checkout week3day2-start My build so far, with the brief in docs/SECURE.md
git checkout week3day2-end Security: an API key for the agent, a check for instructions hidden in claims, and the app ready to run on Lambda
git checkout week3day3-start My build so far, with the brief in docs/CONTEXT.md
git checkout week3day3-end The context layer as its own MCP server, a memory for the agent, and the agent's tools through the Gateway
git checkout week3day4-start My build so far, with the brief in docs/RUNTIME.md and the runtime/ and agentcore/ folders for deploying the agent
git checkout week3day4-end The agent as an AgentCore app, and Sign off or Send back for a claim waiting on a handler
git checkout week3day5-start My build so far, with the brief in docs/COPILOT.md
git checkout week3day5-end The finished product, with the UnderCover Copilot, a chat for handlers in the dashboard's sidebar

About

No description, website, or topics provided.

Resources

Stars

0 stars

Watchers

0 watching

Forks

Releases

Packages

Contributors

Languages