The evidence layer for technical hiring
Every claim on the resume, checked against the work.
Grounded Hire reads each technical resume, finds the public work behind it, and shows your team what is backed by evidence. No scores. No auto-reject. A person decides.
- Evidence, never a score
- Designed for GDPR and the EU AI Act
- Candidates can respond
As written on the resume
Built and ran a payments ledger service in Go
Claim 1 of 6
The inbound channel, in numbers
244
applications per job
The average opening drew 244 applications in 2025, up from 116 in 2022.
91%
of US recruiters
In a 2025 Greenhouse survey, 91% of US recruiters said they had spotted candidate deception.
1 in 4
profiles by 2028
Gartner predicts that by 2028, one in four candidate profiles worldwide will be fake.
300+
US companies
One US facilitator helped North Korean IT workers obtain remote IT jobs at more than 300 US companies.
The problem
Hiring teams stopped trusting resumes. Good candidates are paying for it.
When every resume reads well, none of them tells you much. Teams fall back on the people they already know, and the inbound pile goes unread.
01
Too many to read
Applications per opening more than doubled in three years. Offer rates for inbound applicants fell from 7 in 1,000 to 2 in 1,000. Reading every resume stopped being possible, so many teams stopped reading inbound at all.
02
Referrals became the only door
Referrals are 1% of applications, yet 40% of referred candidates reach interview. An engineer with real work behind them and no contact inside has almost no way through.
03
Some candidates are not who they say
Fake profiles and AI-assisted interview cheating are rising, mostly in remote roles. A polished resume is cheap to produce. A year of commits is not.
“Hiring managers repeatedly tell us they no longer trust what they read in resumes, and that some candidates even turn out to be fake.”
“We are a small company and gave up on inbound hiring. We got too many AI applications.”
How it works
From a pile of PDFs to evidence you can read in a minute.
Grounded Hire does the looking-up a careful recruiter would do if they had an hour per resume, and shows its working.
01
Upload
Drop in a batch of resumes as PDFs. Add a GitHub handle if you have one. If not, Grounded Hire looks for it in the resume.
02
Extract
Grounded Hire pulls out the technical claims as the candidate wrote them, along with the links they chose to include.
03
Collect
It gathers the candidate's public work: which repositories are theirs, how the work built up over time, and what the code is made of.
04
Verify
Each claim is matched to evidence. Every citation is checked to exist before it reaches the report. Statuses are set by tested rules, not by a model's opinion.
05
Ask
Where nothing public exists, you can ask the candidate. They can link other work, name a referee, or describe a private project.
Three questions, asked of every repository.
Each check is tested against resumes and repositories with known answers before it ships.
Is it theirs?
The repository exists and belongs to the candidate. Forks and copies are shown as what they are.
Is it sustained?
Personal commits over months look different from a single bulk upload. Grounded Hire shows the shape of the history.
Does it match?
Languages and dependencies in the code are compared with the technologies the resume claims.
Sample report
What your team actually sees.
One line at the top, then the detail behind it. Open any claim to see the repositories, the history and the technologies that back it.
Evidence summary
3 of 6 claims backed by public evidence
Sample Candidate|github.com/sample-candidate
This is evidence, not a score. Missing public evidence is common for private work and says nothing about ability.
Claim, as written
Built and ran a payments ledger service in Go
An original repository owned by the candidate shows sustained work in Go, and its dependencies include gRPC and a PostgreSQL driver.
Technologies
Go is 91% of the code. go.mod lists grpc and pgx.
Double entry ledger service with gRPC API
Ownership
Original repository owned by the candidate. Not a fork.
History
212 commits by the candidate across 17 active months. Busiest single day holds 4% of them.
Dependencies seen: grpc, pgx, prometheus
When evidence is missing
A missing link is a question, not a verdict.
Most engineering work happens in private repositories. When Grounded Hire finds nothing public for a claim, you can ask the candidate in one click, in your company's name. Their answer lands on the same record.
Link other work
A different GitHub account, an organisation they contributed to, or a specific repository Grounded Hire did not see.
Name a referee
Someone who worked with them on the project and is willing to say so.
Write a few lines
What they built, what they owned, and what was hard. Useful context for private work.
Principles
Rules we do not bend.
Software that touches someone’s livelihood has to be careful. These are built into how Grounded Hire works, not added as settings.
Evidence, never a score
Grounded Hire does not rank, rate or grade anyone. It shows what was found and where, with links you can open.
A person decides
There is no auto-reject and no threshold. The report informs a recruiter. It does not replace one.
Neutral by wording
“No evidence found” means exactly that. Most good engineering work is private, and the report says so.
Private by design
Grounded Hire keeps what the report needs and nothing else. Candidates get a privacy notice, and a candidate can be deleted on request, along with everything derived from them.
Built for oversight
Candidate evaluation is classed as high-risk under the EU AI Act. Every step of every run is logged, so you can show how a report was produced.
Untrusted by default
Resumes and repositories are treated as data, never as instructions. Text hidden in a PDF cannot change a status, because code sets the status.
Where the idea comes from
Paper2Agent, from researchers at Stanford, turns a scientific paper and its code into tools an AI agent can use, and accepts nothing that fails to reproduce the paper’s own results. Grounded Hire applies the same discipline to hiring: a claim counts when the evidence behind it can be shown.
Questions
Straight answers.
Does Grounded Hire score or rank candidates?
No. It reports which claims are backed by public evidence, which are partly backed, and which have none. What to do with that is your call.
What about strong engineers with no public code?
That describes a lot of strong engineers. “No evidence found” is neutral, and the report says private work is the usual reason. You can ask the candidate for a referee or a short write-up, and their reply is attached to the record.
What does it check today?
Public GitHub work: whether repositories exist and belong to the candidate, whether the commit history is sustained and personal, and whether languages and dependencies match the claimed technologies. More sources will follow.
Can someone game it?
They can try, and the usual tricks are visible. A forked project shows as a fork. Code uploaded in one go shows as a single upload. Grounded Hire also lists the signals that tie a GitHub profile to the resume, so you can judge for yourself.
How is candidate data handled?
Resumes stay in your workspace. Only the text needed to extract claims is sent to the language model. You can delete a candidate and everything derived from them in one action, and candidates can ask for the same.
Does it plug into our ATS?
Not yet. Today you upload resumes and read reports in Grounded Hire. A direct connection that writes the evidence summary back to the ATS record is the next step.
Read your inbound applications again.
Upload a batch of resumes and see which claims have work behind them. It takes a few minutes to find out who you have been missing.