Professional Credentialing · AI / ML Engineering
The Verification Standard
for AI/ML Intelligence.
You spent months building something real. Someone else spent a weekend prompting ChatGPT. Right now, no one can tell the difference. Orcred can.
Everyone has a project.
Yours is different.
Right now no one can tell the difference between someone who spent months building something real and someone who spent a weekend prompting ChatGPT. That gap is costing real builders their careers.
1 signal in 24.
That's the problem.
45 minutes.
One engineer. Your work.
Not a quiz. Not a take-home test. A real conversation — about the project you built, every decision you made, every tradeoff you chose. The ones who built it for real talk about it differently.
One session · One verdict
Now there's
proof.
An Orcred Score. A verified credential backed by a senior engineer's sign-off. Something you carry into any room and say — a real engineer reviewed this work. It passed.
Orcred Score
Founding Cohort · 2026
Senior ML Engineer · Verified
They Submit
The project and the person.
A real project. A written explanation of every decision — why they chose this architecture, what they tried that failed, what they'd do differently. No templates. No guided prompts.
Engineers who built something real write differently. Specific. Confident. Uncertain in the right places. The submission is already a signal.
RAG Pipeline · LangChain + Pinecone
Custom retrieval + cross-encoder reranking
Late chunking over fixed-size splitting
Naive top-k at 400-token fixed chunks
120ms latency accepted for accuracy gain
You Talk
45 minutes. Your judgment.
No script. Just you and the candidate and the work they say they built. You go wherever your instincts take you — the tradeoff they glossed over, the decision that seems too confident.
That instinct you've trusted in every technical interview — the one that fires within 60 seconds — now it has a formal home.
Walk me through why you chose this embedding model over ada-002.
It outperforms ada-002 by 14% on domain-specific retrieval in our internal evals. I ran three separate benchmarks before committing.
What would you change if you rebuilt this from scratch?
I'd decouple the chunking pipeline from indexing entirely. Right now they're tightly coupled and painful to iterate on independently.
You mentioned a latency tradeoff — walk me through that decision.
We accepted 120ms over 40ms because the accuracy delta was 18 points on our eval set. The use case justified it.
The Score Stands
Pass or fail. Forever.
An Orcred Score they carry into every room. Backed by your sign-off. Something they can show any hiring manager and say: a real engineer reviewed this work and it passed.
Not everyone passes. That's what makes it mean something.
Senior ML Engineer · Verified
The Orcred
Score.
Not a grade. A signal.
Four things that actually matter.
Technical Depth
Did they build something that works — and do they know why it works?
Communication
Can they walk a room through their decisions without notes?
Reproducibility
Is the work clean enough that someone else could pick it up tomorrow?
Originality
Did they think, or did they follow?
Not everyone passes. That's the point.
Everything else shows.
Orcred proves.
| What it proves | What it misses | |
|---|---|---|
| GitHub | You pushed code. | Whether you understand any of it. |
| LeetCode | You can memorise patterns. | Whether you can engineer a real system. |
| Certificates | You watched the videos. | Whether you can apply any of it. |
| Orcred | You understand what you built. | Nothing. The gap is closed. |
One conversation changes the signal permanently.