Forward Deployed Engineer Portfolio: Projects That Get You Hired

What a portfolio must prove

An FDE portfolio has one job: convince a hiring manager you can take an ambiguous problem to a working, handed-over system. That means each project should show the full arc:

1. The problem in the customer's terms, with a number attached ("support team spent ~6 h/week triaging misrouted tickets").
2. Discovery artifacts - a short brief, success criteria, what you decided not to build.
3. Working code - real error handling, tests, and a README a stranger can run.
4. Deployment story - how it ran in a production-like environment: config, secrets handling, monitoring, rollback.
5. Honest measurement - what improved, what didn't, what broke.

Five portfolio projects that map to real FDE work

1. The messy-data integrator

Take two deliberately inconsistent public datasets (or generate conflicting fixtures), build an idempotent pipeline that reconciles them, and publish the error report format you designed. Proves: data fluency, idempotency, operational thinking.

2. The permission-aware assistant

Build a small retrieval system over documents with role-based access control, then - the important part - write up the tests that tried to leak unauthorized documents and failed. Proves: RAG/LLM competence and security instinct, the combination employers are desperate for.

3. The flaky-upstream survivor

Wrap a real public API (or a chaos-proxy you write) that fails 15% of the time: retries with backoff, idempotency keys, dead-letter queue, and a runbook. Proves: production reliability thinking.

4. The human-in-the-loop workflow

An approval-gated automation: AI drafts, a human approves, an audit trail records who approved what when. Proves: agent design maturity - you know where the human sits and why.

5. The "AI was the wrong tool" case study

Pick a task where a rules-based solution beats an LLM (deterministic classification, exact calculations), build the rules version, and compare cost/latency/accuracy honestly. Proves: judgment - the rarest signal in AI-era portfolios.

Avoid: tutorial clones (todo apps, streaming ETL re-implementations), portfolios of only notebooks, and any project whose README says "demo purposes only" without saying what production would require. Hiring managers read the README first and the code second.

Documenting evidence without exposing data

• Use synthetic or public data; say so explicitly.
• Show fragments of real production incidents if NDA allows (log shapes, not values).
• For each project include a "what I'd harden before real production" paragraph. This single habit reads more senior than any framework choice.

Presenting it

• One page per project, consistent structure (problem → brief → build → deploy → measure → learn).
• Link the repo, but lead with the write-up; make the first 90 seconds self-sufficient for a skim.
• Record a 3-minute walkthrough video if you can - FDE hiring is communication-heavy, and video is the cheapest proof.

How this maps to assessment

Every one of these projects mirrors a deliverable in our practice tasks and the public FDE Practitioner capstone briefs. The rubrics are for self-review; external capstone review is not currently available. Use them as one quality reference alongside the criteria in your target role's hiring process.