The top 1% of forward deployed engineers.
Every engineer we put forward has solved a real problem in our sandbox, working alongside an AI agent. We analyse how they think, build and verify, so you hire on evidence.
sync/worker.py and the last 3 failed runs in logs/. Don't edit yet, just tell me where records could be lost.grep 3 files · read only- 00:04Plan before code · split the brief into 4 steps
- 00:19Caught agent error · rejected an unsafe retry loop
- 00:31Assumption · didn't confirm the SLA with the client
A real-time sandbox, not an interview.
Candidates get a working environment, a realistic brief and our agent. We watch how they actually work, from the first question to the final handover.
The operations team at a large insurer says 2–3% of claims never reach the reporting database. They don't know why, and the original authors have left.
- Find where records are lost, with evidence
- Ship a fix that's safe to run on production data
- Explain the cause and the risk to their ops lead
Six things every great FDE does. Scored from how they work.
Every candidate on your shortlist comes with a profile like this one. Each score is backed by what we observed in the session and the prompt that shows it.
Plans before prompting, keeps the agent on a short leash and never ships what it hasn't tested.
- D14.5
Framing
Scopes an ambiguous brief before writing code.
6 minplanning before first editdon't edit yet, just tell me where records could be lost
- D24.0
Agent direction
Gives the agent clear intent, context and limits.
82%prompts with explicit constraintsuse the claim ID as the idempotency key
- D35.0
Verification
Treats AI output as a draft until it's proven.
9test runs · 2 suggestions rejectedwrite a failing test for the truncation first
- D44.5
Production judgment
Thinks about security, failure and data safety.
3risks raised without being askeda bare except would hide auth failures
- D54.0
Recovery
Spots when things go wrong and changes course.
4 minfrom failing run to diagnosispaste the stack trace, that's not a timeout
- D64.5
Client clarity
Explains trade-offs so non-engineers can decide.
1handover note the ops lead could act ona rerun would duplicate the claims that went through
The AI drives. Output goes in unchecked.
Good on clear tasks. Struggles when the brief is messy.
Directs and verifies the agent. Dependable delivery.
Sets the strategy and catches what the AI misses.
From one call to a shortlist in three weeks.
- 01
Brief us on the role
A 30-minute call and a job description. We agree what the hire will own, and the brief their assessment is built around.
- 02
We source and screen
We map engineers who have shipped production systems for customers, then run a short written scoping exercise.
- 03
Live sandbox assessment
Shortlisted engineers spend 60–90 minutes in the sandbox with our agent. We score the whole session, and a person reviews it.
- 04
Meet the top 1%
You receive a handful of candidates, each with a scorecard and the session evidence behind it. You interview and make the offer.
Pay only when you hire.
No platform licence and no retainer. You pay a placement fee when a candidate we introduce accepts your offer.
Discuss terms- No retainer or upfront fee
- Fee charged only on a successful hire
- Volume terms for multi-role and GCC build-outs
- One agreement for hiring across regions
Questions, answered.
Something else? Email nayan@fdefactory.in
No. Nothing connects to your Slack, code, ATS or internal tools. We start from a conversation and a role description, and you receive candidates and their reports by email or a secure link.
Tell us who you're hiring.
Book a 30-minute call
The fastest way to start. Tell us about the role and the team, and we'll show you a sample assessment report.
Pick a time