Goal: Land a Forward Deployed Engineer (FDE) role at an AI startup / lab.
Owner: Mark Shcherbakov (Lowcoding.dev) · Started: 2026-07-20
How to use: Work skill-by-skill. Each skill lists topics and resources. Priority tags tell you where your gaps actually are — spend time there, skim the 🟢 strengths.
| Tag | Meaning | Why |
|---|---|---|
| 🔴 Focus | Real gap — most of your time here | Solo low-code/AI work doesn't build this |
| 🟡 Level-up | Partial — sharpen it | You touch it but not to FDE bar |
| 🟢 Strength | Polish only | 6yr client delivery + CPO already cover this |
Your edge: end-to-end solo delivery, real enterprise clients (Miro, JTI, SDG, Next Energy, Rebel Fund), AI-native stack (n8n multi-agent, Supabase, MCP, LLM agents), CPO-level product+GTM sense, days-not-weeks prototyping.
The delta to close: traditional SWE rigor · agent/eval rigor over vibe-checks · enterprise system design (SSO/VPC/compliance) · the case/decomposition interview.
FDE coding rounds are integration/debugging/production-quality code — not LeetCode. But you need to write idiomatic Python in a real codebase, fast and clean.
| Topic | Priority | Resources |
|---|---|---|
| Idiomatic Python (stdlib, typing, dataclasses, async) | 🔴 | AI Eng from Scratch — Python track, [Fluent Python (book)] |
| Data structures & algorithms (light — enough to reason, not grind) | 🟡 | NeetCode 150 (skim patterns) |
| Git, testing (pytest), CI, code review culture | 🔴 | pytest docs, GitHub Actions basics |
| Reading & debugging unfamiliar/large codebases | 🔴 | Practice on SWE-bench Verified tasks |
| API integration + data transformation/pipelines | 🟡 | Build against real 3rd-party APIs (you already do this) |
The core of an AI FDE. You build agents already — the gap is rigor: evals, reliability, structured tool-use, cost/latency discipline.
| Topic | Priority | Resources |
|---|---|---|
| LLM API mastery: tool-calling, structured output, streaming | 🟡 | Anthropic docs — tool use, OpenAI cookbook |
| Prompt engineering & context design | 🟢 | Anthropic prompting guide |
| Agent loops, planning, memory, tool orchestration | 🟡 | AI Eng from Scratch — agent phases |
| MCP server/client design | 🟢 | You already do this — document it as a portfolio piece |
| Evals & observability (the real gap) | 🔴 | Anthropic evals guide, build an eval harness for your agent |
| Reliability: retries, guardrails, failure handling, pass^k | 🔴 | τ-bench paper/method |
| Cost / latency / token tradeoffs | 🟡 | Instrument your own runs |
A sub-skill of agents, but big enough to isolate. Directly tested in AI-FDE system design.
| Topic | Priority | Resources |
|---|---|---|
| Embeddings, chunking, vector DBs (pgvector/Supabase) | 🟡 | Supabase pgvector |
| Retrieval quality, hybrid search, reranking | 🔴 | Enterprise RAG Challenge (warm-up capstone) |
| Query routing, expansion, self-consistency | 🟡 | ERC winning write-ups (Ilya Rice) |
| Citations, grounding, hallucination control | 🔴 | Build "chat with docs" + citations |