Roadmap

Learn Forward Deployed Engineering

6 phases and 11 skills, from strong engineer to Forward Deployed Engineer. Each phase ends with a project that proves the skill. Click any skill for the full guide.

The roadmap

1. 1

   Engineering baseline

   2–4 weeks · skip if you ship production code today

   Be able to build, deploy and run a small app on your own, and explain what happens when it breaks.

   Practical coding & data

   You're ready to move on when

   • You write Python and SQL with tests, logging and config instead of hardcoded values.
   • You can deploy to a live URL with environment variables and no secrets in the code.
   • You're comfortable with Git, the command line and calling REST APIs.

   Proof-of-skill project

   A deployed app with a failure plan

   Ship a small CRUD app with login to a public URL. Its README explains how to run it, what happens when the database or an API is down, and how you'd roll back a bad deploy.

2. 2

   AI foundations & first builds

   4–6 weeks

   Understand how LLMs behave and build something that makes useful decisions from messy input.

   Applied GenAI: RAG, agents & toolsEvaluation & quality

   You're ready to move on when

   • You can explain tokens, embeddings and context windows without jargon.
   • You use structured outputs and write prompts as precise specs.
   • You measure quality on a small real dataset instead of eyeballing it.

   Proof-of-skill project

   Triage app from a three-sentence brief

   Take a vague brief ("help our team sort incoming requests"). Write category definitions, measure how often two people agree on 30 samples, build a classifier with structured output, keep keys server-side, and ship it to a live URL with its accuracy on your test set.

3. 3

   Discovery & consulting core

   4–6 weeks · run alongside the next phase

   Diagnose before you prescribe. Learn to find the real problem and get people aligned around it.

   Problem discovery & reframingWorkflow & process mappingStakeholder communication

   You're ready to move on when

   • You turn vague requests into sharp problem statements with a baseline and a trust threshold.
   • You map real workflows, including exceptions, and spot the decision points.
   • You can write and present an answer-first recommendation in two minutes.

   Proof-of-skill project

   One-page discovery memo + two-minute pitch

   Pick a realistic business with several problems (a clinic, a logistics firm, a support team). Rank the problems with evidence, recommend one pilot, name its owner, list what you won't build, then record a two-minute executive pitch.

4. 4

   Production GenAI & integration

   8–10 weeks

   Build systems that survive real data, real permissions and real failures.

   Applied GenAI: RAG, agents & toolsEvaluation & qualityIntegration & enterprise systemsSecurity, identity & data governanceReliability & incident response

   You're ready to move on when

   • Your RAG system filters by permission before retrieval and cites its sources.
   • Every change runs against an eval suite with written release gates.
   • Your ingestion job is retry-safe and you can see every request in traces.

   Proof-of-skill project

   Permission-aware document assistant

   Build an assistant over a document set with per-user permissions. Include citations, an eval suite with gates, request tracing, a retry-safe incremental sync, a written failure register, and a short security data-flow diagram.

5. 5

   Client simulation & ownership

   4–6 weeks

   Perform under pressure with real people in the loop: pushback, incidents and adoption problems.

   Stakeholder communicationReliability & incident responseRollout & adoptionOwnership & business value

   You're ready to move on when

   • You handle a security block, an impatient exec and an angry user in role-plays without caving or arguing.
   • You can run the first 90 minutes of an incident and write the customer update.
   • Your rollout plans have gates, stop conditions and named owners.

   Proof-of-skill project

   End-to-end mock engagement

   With a partner playing the customer: run discovery, build a small pilot, deploy it, present it, then handle a simulated incident and write the postmortem. Record it and review it against the interview rubric.

6. 6

   Interview readiness & senior judgment

   2–4 weeks · repeat as needed

   Turn skill into offers, and show the judgment that separates senior FDEs.

   Problem discovery & reframingPractical coding & dataOwnership & business value

   You're ready to move on when

   • You run a scenario answer framework by reflex: context, current system, success, integration, failure, security, rollout, trade-offs.
   • You have 6–8 first-person stories with numbers, covering ambiguity, incidents, conflict and delivery.
   • You score 4/5 or better on every rubric dimension in three unaided mocks in a row.

   Proof-of-skill project

   45-minute capstone, no notes

   Take a struggling AI pilot that users have stopped trusting. In 45 minutes, diagnose it, propose a fix, define evals and gates, plan the rollout and write the executive update, then score yourself honestly.