How startups hire stellar junior talent in the age of AI slop code.
Hire tech builders, proven on the work they ship.
Sixteen built it. You read the work and interview the top.
“Cut the dashboard to ship the pricing page on time, then hand-wrote the experiment hook to own the logic.”
“Clean pipeline, but leaned on the model for the tests.”
“Shipped it, but let the scope creep past the brief.”
Great products are not made by engineers who close tickets, but by Builders.
A short walk through the product, from the job description you paste in to the scored work that comes back.
How do you know whether they can code well with AI?
A clean CV, a Claude-ed take-home and a rehearsed interview all come out of AI models now, so you can't assess junior engineers from AI chaff.
The juniors who steer AI well ship like seniors, which makes them the highest-value hire on the market. You can reach them before anyone else does.
11,000 job applications a minute on LinkedIn, up 45 per cent in a year.
LinkedIn, reported by The New York Times, 2025
By 2028, 1 in 4 candidate profiles will be fake.
Gartner, 2025
Replacing an employee can cost 50 to 200 per cent of their salary.
SHRM, 2022
Three steps. You do the first and the last.
Turn the role you've already written into a project builders make for you.
Founding engineer. Own features end to end, from a loose problem to something shipped this week. Fast with the model, strong product instincts, comfortable across the stack.
Chose BM25 over embeddings. The corpus is small and exact terms matter more than fuzzy matches. Benchmarked both and kept the loser in the notes.
Capped results at k to hold latency under 50ms, then tested the worst-case query first.
Post the role you already have.
Paste the job description from LinkedIn, your ATS or a Google Doc. Candela reads it and drafts the project a strong builder would ship to prove they can do the job: three milestones, starter files, and the rubric the scorer will use. Edit any of it, publish, and it goes live. The people who choose to build it are self-selecting into your role.
You · five minutes
Founding engineer. Own features end to end, from a loose problem to something shipped this week.
Builders pick it up and reason out loud.
As they work, each commit carries a short reasoning log beside the code: the trade-off they made, what they ruled out, and why. You read how someone thinks while they build, not just what they handed in at the end.
Candela · one week
Chose BM25 over embeddings. The corpus is small and exact terms matter more than fuzzy matches. Benchmarked both and kept the loser in the notes.
Capped results at k to hold latency under 50ms, then tested the worst-case query first.
Browse the wider cohort.
You're not limited to the people who started your role. Search the whole cohort in plain language, like “shipped a Python data tool and can explain the trade-offs,” and shortlist anyone whose work you want to read more closely.
You · 90 seconds a read
How builders get assessed, in five dimensions.
When a build ships, Candela scores it into a credential, a scorecard across five dimensions, each carrying one sentence of quoted evidence pulled from the builder's own reasoning log, milestone reviews and file tree. You read what the work shows, with the line that proves it sitting right next to the score.
Reasoning depth
How honestly they weighed the trade-offs in the reasoning log.
Scope discipline
What they shipped, measured against the brief they were given.
Iteration honesty
Whether the dead ends are written down, or buried.
Code quality
Structure, tests and error handling, read at code-review depth.
Communication
Whether a stranger can understand the work in five minutes.
Each dimension scores out of five.
The project gets an overall percentage and a band.So when you open a builder, you might see an 84% that weighed three options for a retriever and benchmarked why it rejected one, kept all three milestones inside the original brief, and shipped a README that explains the work in a paragraph. That's the read you used to get an hour into an interview, available before the first call.
One hire who ships product, talks to customers and runs the demo.
We call them a Founder's Builder.
A technical graduate who builds with Claude well enough to ship product end to end. The rest they learn on the job.
What you get that a CV and a take-home can't.
Verified before you call
The credential shows how a builder thinks and where they made the hard calls, each line backed by quoted evidence, so you judge the substance rather than a polished application a model could have written.
Trained before you meet them
Builders come through several deployed projects, coached the whole way by a system that questions each decision until they can defend it. The cost of bringing someone up to speed sits with us, not your senior engineers.
You pay on the hire
Your hiring managers get platform access, and a placement fee only lands when someone joins.
Full pricing is walked through on the call.
The cohort is live, and builders are shipping scored work.
We started with builders at Imperial College London, and we're widening across the top UK technical universities.
“I left Imperial with four job offers, all of them from the projects I could open on a laptop and walk a founder through. Most graduates I know never get that moment, partly because they haven't built the work and mostly because no one has given them a way to show it. Candela exists so the next junior with the work can walk into the room with it already on the table.”
If you're the one building, start here.
Candela is free if you're the one writing the code. Pick what you want to build, get coached through it, and finish with projects a hiring manager can open and run. The startups above read your work, not your transcript.
Start building →Assess Tech Cracked-ness through Real Work not AI CVs.
We'll set you up with access to the cohort, and walk you through a scored project the way you'd read one before an interview. Reading one takes ninety seconds.
Prefer to look first? See a scored project