Sample Scorecard
One scorecard per candidate. Every shortlist.
A real example of the AI hard-skills report we deliver as part of every shortlist. Structured evaluation, per-question scores, English assessment, and a clear recommendation — reviewed by a human recruiter before delivery.
Interview Details
Oleksandr K. — Senior ML/LLM Engineer
ID: 6f842308-4290-4433-ade2-9be***
Candidate Information
Name: Oleksandr K.
Email: o.k***@gmail.com
Status: completed
Created At: 3/3/2026, 9:53 PM
Position Details
Role: Senior ML/LLM Engineer
Field: AI & Machine Learning
Type: With Manager
Final Recommendation
Candidate demonstrates strong applied LLM engineering depth across RAG architecture, eval design, and retrieval tuning with an average score of 8.6/10. Shows excellent judgement on fine-tune vs prompt engineering decisions and a systematic approach to hallucination and regression testing. Minor gaps include limited hands-on distributed fine-tuning experience and could quote more concrete eval metrics from past projects. Current salary $70K, targeting $75-85K — within range for a senior ML engineer in CEE.
English Assessment
Communication & clarity signal
A1
A2
B1
B2
C1
C2
Key takeaway
Communicates technical concepts clearly and fluently. Can lead architecture discussions and explain complex tradeoffs to both technical and non-technical stakeholders. Minor grammatical imprecisions that do not impede understanding.
Question-by-Question Evaluation
Overall Performance Summary
The candidate demonstrated strong applied knowledge across all LLM engineering topics with consistent performance scores of 8-9. Technical understanding was evident throughout, with particularly strong answers on RAG architecture and eval design.
General Feedback
The candidate shows excellent practical knowledge of retrieval pipelines, eval methodology, fine-tuning decisions, and production LLM monitoring. They consistently identified key tradeoffs and appropriate solutions across all areas. Their responses demonstrated real-world experience shipping LLM features to production.
Strengths
Demonstrates comprehensive practical knowledge of RAG architecture, eval design, and retrieval tuning with strong awareness of production-readiness concerns like regression suites, canary rollouts, and cost-latency budgets.
Areas of Improvement
Should provide more specific examples from past fine-tuning work, particularly around distributed training and LoRA hyperparameter selection, and could quote more concrete metrics from previous eval projects.
Overall Assessment
The candidate shows senior-level working knowledge suitable for an ML/LLM engineering role with demonstrated ability to reason about retrieval, eval, and deployment tradeoffs at scale.
Individual Question Evaluations
Candidate Answer:
The first trade-off is chunking: smaller chunks improve retrieval precision but lose context, so I'd start around 512 tokens with overlap and tune against a labelled query set. For retrieval I'd go hybrid — BM25 plus dense embeddings — because pure vector search misses exact terms like product codes. At 2M documents you need a reranker: retrieve top-50 cheaply, rerank to top-5 with a cross-encoder. Then there's the context budget — stuffing more chunks raises cost and can hurt answer quality, so I'd measure groundedness against context size. Finally, incremental indexing for freshness rather than full re-embeds.
Evaluation:
Excellent RAG design answer covering all critical components. The candidate correctly framed chunking, hybrid retrieval, and reranking as measurable trade-offs rather than fixed recipes, and anchored each decision to evaluation against a labelled query set. The mention of incremental indexing and context-budget cost shows practical experience running RAG at scale.
Key Takeaway:
Strong RAG architecture skills with clear reasoning about retrieval precision, reranking, and cost tradeoffs. Production-ready thinking.
This is the AI hard-skills report we deliver per candidate, alongside the recruiter soft-skills notes and CV analysis. Generated as part of our 6-step shortlist process — every shortlist is signed off by a human recruiter.
Assess the skills. Review the signals.
Anti-cheating signals help our recruiters investigate concerns before candidate introductions. For roles that use AI tools, an optional coding assessment also examines how candidates work with them.
AI changes the skills worth assessing
GitHub Copilot, Cursor, ChatGPT — AI-assisted development is the default. But most hiring pipelines still test for pre-AI skills only.
AI proficiency ≠ copy-pasting prompts
The best engineers know when to use AI, when to override it, and how to validate its output. That judgment is what the assessment should make visible.
Bad AI habits cost more than no AI at all
Blindly trusting AI-generated code leads to security vulnerabilities, hallucinated logic, and tech debt that compounds silently for months.
AI-assisted coding assessment
An optional module added to the included technical assessment: real coding tasks with AI tools available, followed by probing questions about the candidate's AI-assisted workflow. The result is a separate AI Proficiency Score alongside the standard technical evaluation.
How it works
The candidate gets access to Copilot/Cursor during part of the interview. We observe how they prompt, validate, and iterate — then score their AI fluency on a structured rubric.
AI-assisted coding
Can they use Copilot/Cursor effectively while catching errors?
Prompt engineering
Do they write precise prompts, or do they brute-force trial and error?
AI output validation
Can they spot when an LLM hallucinates, introduces a vulnerability, or produces subtly wrong logic?
AI-native architecture
Do they know when to reach for an LLM vs. a deterministic solution?
Recruo Secure Browser
Our interview environment adds controls and session-integrity signals to support review. The available controls depend on the setup; no browser can prove that outside assistance was impossible.
Why it matters in 2026
A polished answer does not establish practical experience. Adaptive follow-ups and evidence review help recruiters understand how the candidate reasons and which claims need clarification.
Locked browser session
Ask which browser controls are supported in the agreed interview setup.
Copy-paste detection
Available session signals can highlight material that needs follow-up.
Recruiter follow-up
Concerns are reviewed in context, with clarification from the candidate where needed.
Evidence and context
Review the answers and session evidence together; a flag does not prove external assistance.
Assessment options are scoped around the role. Available as add-ons for Standard searches. For Executive Search, the leadership assessment format is agreed privately with sales.
Most used when hiring LLM engineers, RAG engineers, Evals engineers, ML platform engineers, Agentic AI developers, Generative AI engineers or AI/ML engineers.
Get started
Get a scorecard like this for your candidates
Scope your role and agreed assessment stages. Meet finalists with assessment evidence and confirmed expectations. Standard: 18% total, $2,000 credited at launch; balance when your hire starts.
$2,000 to launch
Fully credited toward the 18% Standard fee
90-day guarantee
One replacement search under agreed conditions
Human review
Recruiters review the screening evidence