Stop reading. Brief me.
This page is a working Forward Deployed Engineer simulation. Give me your real, vague, messy problem. I'll run the FDE "decomposition" interview on it live: scope it, draw the architecture, plan the sprint, and call out where it'll fail. Then map every phase back to projects I've shipped.
Receipts
The simulation above isn't vibes. Here's the engineering it runs on.
Four proofs of the range FDE work needs: agents, protocols, upstream code, distributed systems.
Vague spec → working multi-agent in 10 days.
Multi-agent LangGraph research system, built from a rough problem statement. Hexagonal architecture, verification loops, local + cloud LLM providers, swappable by config.
Made my own site agent-queryable via WebMCP.
7 tools in production. Early implementation on the exact surface Anthropic and Google FDE postings now call table-stakes.
Walked into unfamiliar code and left it better.
Merged PRs into the Anthropic MCP Python SDK and into vLLM (200k+ LOC), where the PR added LoRA support for LLaVA's tower and connector. The exact muscle FDEs use in customer code.
Services, data flows, contracts, failure boundaries.
Kayak: 3-tier distributed architecture. Node/Express services behind API gateway, polyglot persistence (MySQL/Mongo/Redis), Kafka, FastAPI AI layer. Airbnb on Kubernetes microservices.
Candid
Notes on fit.
The version where I'm honest about what I can claim, and what I can't. Yet.
What I can credibly claim
The engineering: agentic systems, protocols, upstream code, distributed services. The decomposition muscle the simulation above shows.
Plus real stakeholder-facing delivery experience: requirements alignment, metric and SLA definition with finance and operations at Elite Hotel Group.
What I haven't yet
The full FDE customer lifecycle in an external environment. Internal stakeholder delivery isn't the same as external customer delivery. I won't pretend otherwise.
I'm closing this by shipping one small real deployment, publishing failure analyses, and turning an existing project into a deployment case study. Specifics on request.
If the simulation made you think, say so.
Fastest path: email. I read every one. If you ran the sim on a real problem and it sparked an idea, send me the brief and I'll show you what the next 30 minutes of work would look like.