Landing the Role

⏱ ~9 min read · 🎧 10 min listen · ✎ 3 quizzes · 🧪 ~90 min lab

This is optional track A. If you're here to land an FDE role, this lesson is your playbook — work it fully. If you took this course purely to upskill, skim it and move to Lesson 13; nothing downstream depends on it. And if you're not sure yet: do both tracks. You lose nothing, and a published portfolio is worth having even if you never interview. Here's the shift this lesson makes: for eleven lessons you've been building capability. Now you package it as evidence — because FDE loops don't test whether you can learn this job. They test whether you can already show it.

The loops, decoded

The good news about FDE interviews in 2026: the major loops are documented and they rhyme. Exponent's guide — the closest thing to a canonical source — breaks down the three that matter most.¹ Palantir runs the longest loop, up to seven stages, starting with a Karat-administered coding screen and building to its signature decomposition round. OpenAI runs recruiter screen → a take-home budgeted around five hours → a walkthrough of what you built → a three-to-four-hour onsite. Anthropic runs a take-home, a 90-minute coding interview, and a mission-alignment conversation.¹ Same species throughout: real-world building plus consulting judgment, thin on LeetCode-style puzzles.

The take-home: scope beats completeness

The take-home is where strong engineers sabotage themselves, because they play it like a feature sprint: build everything, polish everything, sleep never. What's actually scored — per Paraform's analysis of OpenAI's loop — is scoping and narrated reasoning.² A submission that ships a smaller, working core and includes a README saying "here's what I deliberately cut, here's why, and here's what I'd build next with another day" beats a sprawling half-finished everything-app. That's not interview theater — it's the FDE job itself. Every engagement is a scoping decision under a time budget, and the take-home is a rehearsal of it with the interviewer watching. Then comes the walkthrough, where the artifact matters less than your ability to narrate the decisions inside it: why this architecture, why this eval, what breaks first at scale, what you'd never ship to production as-is.²

The decomposition round: don't touch the solution

Palantir's decomposition round hands you a deliberately vague enterprise problem — "a logistics company wants to reduce delays" — and watches how you structure it. The single most common rejection reason, per Exponent: jumping to a solution before the problem is decomposed.¹ Everything you trained in Lesson 10 is the counter-move: who are the actors, what does the current process actually look like, where's the data, what would "better" measurably mean, which constraint binds first? Only after that skeleton exists do you propose anything — and then you propose the smallest thing that tests the riskiest assumption. This round is problem discovery under exam conditions, and it's trainable: the lab below gives you a timed rep.

Your portfolio: a case-study index, not a code dump

Nobody reads code first. A hiring manager gives your repo ninety seconds, and what they can absorb in ninety seconds is stories with evidence — which is why your portfolio README becomes a case-study index: one section per artifact, each stating the problem, what you built, the eval evidence it works, and a link into the code for whoever wants depth. Your capstone from Lesson 11 is the anchor case study — including its "what we recommended NOT to build" section, which quietly demonstrates the AI judgment hiring managers say is hardest to find. Anthropic's own posting asks for "production experience with LLMs including advanced prompt engineering, agent development, evaluation frameworks"⁴ — your repo should let a reader tick each phrase against a named artifact without opening a single source file.

Targeting: three markets, three trades

Three places hire this shape, and they trade off differently. Frontier labs (Anthropic, OpenAI) pay the most — roughly $300–550K total comp mid-senior, $600K–1.2M at staff level³ — with the highest bar and heavy travel; Anthropic's posting lists $200–300K base with 25–50% travel.⁴ AI-forward enterprises and platforms (Databricks and its peers) pay strong-but-lower comp for a broader mandate across more customers.⁶ Consultancies and services firms are the most accessible entry — genuine customer-facing delivery reps, lighter LLM depth, a stepping stone rather than a destination. The market is expanding across all three tiers as companies discover deployment, not modeling, is the bottleneck.³

Now the honest paragraph for readers in Australia: most frontier-lab FDE roles are anchored to US and European hubs, with travel to customers from there.⁴ A Sydney-based frontier-lab FDE role exists occasionally, not reliably. Your realistic near-term market is: remote-friendly FDE and applied-AI roles at AI-forward companies, the local offices of global platforms, and Australian enterprises building AI capability in-house — banks, health tech, government suppliers — where your virtual-healthcare domain depth is a genuine differentiator, not trivia. Aim applications at that market now, keep frontier-lab loops as the stretch target, and remember the skills are identical — only the letterhead differs.

The resume and the stories

Finally, the framing layer. FDE resumes and behavioral rounds run on engagement stories: STAR-format narratives (situation, task, action, result) where the result is a business outcome, not a technical fact. "Built a RAG pipeline" is a feature; "cut clinician document-lookup from minutes to seconds, verified by an eval suite the customer now runs in CI" is an outcome — and hiring data shows the market has shifted toward exactly this hybrid of technical depth and customer-facing evidence.³ Your course artifacts are legitimate story material: they're simulated engagements, and you say so plainly — honesty about the synthetic setting plus concrete eval evidence beats vague claims about real work every time.

🧪 Practical steps: portfolio publish + interview kit (~90 min)

This lab turns D:\Projects\FDE-Portfolio from a folder of lab outputs into a published, interview-ready asset — the moment eleven lessons of artifacts become hiring evidence. Full instructions are in labs/0012-interview-kit.md; the kit lands in D:\Projects\FDE-Portfolio\l12-interview-kit\ and is tracked in the course's artifact tracker.

1. Polish the portfolio README as a case-study index. One section per artifact: problem → what I built → eval evidence → link. Lead with the Lesson 11 capstone. Pick your public GitHub repo name now — short, descriptive, professional (e.g. applied-ai-field-portfolio).
2. Publish to GitHub — private first, then run the pre-public checklist: no API keys or secrets anywhere in history, no real PII, a clear synthetic-data disclaimer in the README. Flip to public only when all three pass.
3. Write 3 engagement stories in STAR format from your course artifacts (capstone, evals lesson, integration lesson are the richest). Each must end in a stated outcome — a number, a decision, or a shipped thing. Save as engagement-stories.md.
4. Run one timed 30-minute mock decomposition. The lab file contains a prompt that turns your AI assistant into an interviewer holding a vague enterprise problem. Your one rule: decompose aloud or in writing — actors, process, data, success measure, binding constraint — before proposing anything. Save the transcript as decomposition-rep-01.md.

Feedback loop: bring the README, the stories, and the decomposition transcript back to your AI assistant with the review prompt below. This kit is the terminal artifact of the portfolio thread that started with your Lesson 01 gap map — re-score that gap map now and put the before/after in the README; it's an interview story in itself.

You are an experienced Forward Deployed Engineer hiring manager reviewing my interview kit: a portfolio README written as a case-study index, three STAR-format engagement stories from simulated engagements, and a transcript of a timed 30-minute mock decomposition exercise.

Grade each criterion as Strong / Adequate / Missing, with one sentence of evidence:
- README scannability: in 90 seconds a hiring manager can name each artifact's problem, what was built, and the eval evidence — without opening code.
- Story outcome-strength: every STAR story ends in a concrete outcome (a number, a decision, a shipped thing), not a technology list — and is honest that the engagement was simulated.
- Decomposition discipline: in the transcript I mapped actors, process, data, success measure, and constraints BEFORE proposing any solution; flag the earliest moment I started solutioning.

Be skeptical — find the weakest story and interrogate it like a behavioral round. Then tell me the one change to the README that most improves my odds of a callback.

My interview kit follows below.

Check yourself — think like a candidate being scored

Scenarios, not recall. Diagnose from the mental model — don't scroll up. Wrong picks stay live.