AI Hiring
AI Engineer for an Internal AI Copilot
A product company building an internal AI Copilot needed an engineer with real production LLM experience — not a GPT-wrapper portfolio. We ran the search end to end: 1,000+ profiles reviewed, a top-5 shortlist, and a candidate brought to offer twice.
Client
Product company building an internal AI Copilot (under NDA)
Challenge
The market is full of engineers who talk well about AI but have never run an LLM system under real traffic. The client needed someone who could own the Copilot architecture — retrieval, evaluation, latency, cost — within a defined budget of ~$8k/month. The search also ran through peak vacation season, with the CEO, CTO and HRD each away at different times.
Hiring Strategy
Before contacting a single candidate we pressure-tested the role together with the client: what the Copilot must do, what “production” means for this system, and what must ship in the first 90 days. The evaluation bar was set on production evidence — real RAG/LLM systems with users behind them, not demos.
Search Process
1,000+ profiles reviewed, 100+ candidates screened in depth, several dozen interviews. The final top-5 shortlist included graduates of the Yandex School of Data Analysis and MIPT, and engineers from a handful of the largest tech companies on the market. Total timeline: about two months — across vacation season and a full role revision midway.
Result
The first candidate was brought all the way to a signed-off offer — and then couldn't start for personal reasons. We didn't reset the engagement: the search restarted immediately, and within the same process we delivered a second, even stronger candidate. The final offer is in progress right now.
Technologies
- LLM
- RAG
- Python
- Evaluation pipelines
Business Impact
The client never had to restart hiring on their own: when the first offer fell through, the pipeline, the evaluation bar and the market map were already in place — the second candidate came from the same funnel, not from zero.
Key Takeaways
- A signed offer is not the finish line — a hiring partner's job includes what happens when it falls through.
- Shortlist quality beats volume: 1,000+ profiles compressed into five people the client would genuinely hire.
- Production evidence — not interview polish — is the filter that separates AI engineers from AI talkers.
Don't lose months and budget on the wrong AI hire.
A 30-minute breakdown of your challenge. No pitch. We'll show how we'd approach it ourselves — including the option not to hire yet.