What we test for
Typical work
- Retrieval over your own documents, with answers that cite their sources
- Agents and tool calling wired into the systems you already run
- Eval suites that catch a regression before your customers do
- Taking a prototype that works down to a cost and a latency you can ship
The coding test
Build a retrieval endpoint over a small set of documents: chunking, embeddings, a vector store and an answer that cites its sources. Then we hand over twenty questions with known answers and ask for the numbers: where it fails, why, and what they would change first. Scored on the eval, not the demo.
Inside AI
The ones people search for by name get a page of their own, with their own coding test and interview questions.
LLM and RAG
Retrieval over your own documents, with citations and an accuracy number behind it.
See the page →AI agents
Agents that act on real systems, with permissions, budgets and a trace you can read.
See the page →AI QA and evals
Golden sets, evals that gate a release, and red teaming by people who test for a living.
See the page →MLOps
Serving, versioning, drift monitoring and the GPU bill, for models already in production.
See the page →Five interview questions you can reuse
We use questions like these in our soft-skills and technical rounds. Steal them for your own interviews.
The answers look right in the demo and wrong in production. Where do you look first?
A strong answer covers: Retrieval before the model: what actually reached the context window. Chunk size and overlap, the embedding model, whether the query needed rewriting, and whether the right document was even in the index. Swapping the model is the last move, not the first.
How do you know a prompt change made things better?
A strong answer covers: A fixed set of cases with known answers, scored the same way every time, run before and after. Per-case diffs rather than one average, and a human read of the cases that moved. Without that, a prompt change is a guess with a good feeling attached.
A stakeholder wants AI in the product and has not said what for. What do you ask?
A strong answer covers: What decision or task it is meant to change, how you would know it worked, and what happens when it is wrong. If nobody can name the measure or own the wrong answers, the honest advice is that it is not ready to ship.
A user pastes "ignore your instructions" into a field that reaches the model. What have you already done about it?
A strong answer covers: Treated model output as untrusted input: nothing goes straight into a shell, a query, or a tool that holds real privileges. Tools scoped to the task, secrets kept out of the context, and the injection actually tested rather than assumed away by the system prompt.
It works and it costs too much. What do you cut?
A strong answer covers: Measure tokens per request before touching anything. Then a smaller model on the easy path, caching the stable prefix, retrieving less but better, and shortening the output. Batching and a blanket model downgrade come last, once you know which call is spending the money.
One monthly price per developer. Everything in it.
No deposit, no minimum term, no placement fee. You pay one invoice a month for each developer on your team.
Per developer, all-in. Your quote arrives by email with your exact number after you send the brief.
- Developer pay, local payroll and taxes
- Vetting: coding test, soft skills, English
- Free replacement within 90 days
- A direct line to our CEO
What changes the price
- Seniority and role. Senior engineer, tech lead or architect.
- Stack. Some stacks and niche tools cost more to staff.
- Team size. Larger teams get a better monthly rate per seat.
- White-label. Agencies reselling under their brand get partner pricing.
Terms in plain words
- First 2 weeks risk-free: if it isn't working, you don't pay for them
- Month to month, 30 days' notice to end
- No deposit and no minimum number of seats
- Want to hire the developer full-time? Buyout of 3 months' fee in year one, 2 months' after
Questions tech leads ask us
How senior are your AI engineers?
Six or more years as software engineers, and at least one AI feature they took to production and still maintain. A tutorial project does not count, and we can tell the difference, because we ask to see the evals.
Everyone says they do AI now. How is this different?
We publish the exercise before you see a profile, and it is scored on measured accuracy rather than on a demo that worked once. TopDevCrew is part of a QA company, so measuring a system nobody else knows how to measure is the part we are good at.
Can I see the coding test results?
Yes. Every profile card carries the score and the reviewer notes, including how that candidate scored on our question set.
What does it cost?
$6,900 per developer per month, all-in, and flat: the same rate whatever the stack. Your exact quote arrives by email after you send the brief.
Skip the sales rep. Talk to the CEO.
Tell us what you're building, your stack and when you need people. Three matched profiles within 48 hours, and a straight answer on whether we're the right fit.