From 2,000 applicants to 5 evidence-backed finalists, inside your ATS.
Unfound runs everything between the application and the offer: AI screening, a live simulation of the job, a copilot for your interviewers and an evidence report. One flow, one candidate record, no handoffs.
Five tools and four handoffs, or one flow.
Most teams stitch hiring together from separate tools, and every handoff loses context. Unfound runs it all on one candidate record.
Connect your ATS. Keep your sourcing.
Unfound plugs into the applicant tracking system you already run. Applicants flow in from your job posts as they do today, and results land back on the candidate record.
- Works with Greenhouse, Lever, Ashby and Workday
- Your job posts, careers page and sourcing stay as they are
- One candidate record from application to offer
Every applicant read. The shortlist screened.
A smart filter sorts every applicant in minutes against what the role actually needs. AI screening then reviews the filtered shortlist, so your team starts from a list worth reading.
The role’s must-haves, evidence of the work itself (scope, outcomes, what they owned) and the requirements you set for the role.
Names, photos, age and other personal characteristics, school prestige, and polish: a well-written application earns nothing on its own.
The shortlist works the actual job.
The strongest candidates get an invite to a multi-agent job simulation built from the role. AI agents play the client, the engineers and the VP, and the scenario forks around every decision the candidate makes.
The Q3 data contradicts the launch plan. Launch is Thursday.
Your interviewers walk in knowing what to probe.
The copilot sits beside the live call with the simulation replay. It shows the moments that matter and suggests the questions that test them, so the interview checks evidence instead of collecting more claims.
- Replay any decision from the simulation during the call
- Suggested questions, tied to specific moments
- Your interviewer runs the conversation; the copilot never scores the call
“…so I’d usually align the client first, then go back to engineering.”
Probe suggested“At 07:15 you paused the launch. What would have made you ship on Thursday instead?”
Decide on evidence, not gut feel.
The evidence report hands your team a judgment profile, and every score links to the timestamped decision behind it. Compare finalists on the same rubric and replay anything before you trust it.
Hit contradictory data and paused the launch to re-pull it before committing a date.
Candidates use AI to apply. There’s nothing here for it to rehearse.
A prep tool can write a polished interview answer in seconds, but there is no script to memorise here. Every scenario is generated for the candidate and changes with each choice they make, so there is nothing to leak and nothing to rehearse. What counts is the decisions they make while it runs.
Generated per candidate
Each run is built from the role with its own seed. Last week’s candidate can’t pass this week’s scenario along.
Forks at every decision
Five decisions open 32 possible worlds. The agents react to what the candidate just did, so a memorised answer has nothing to attach to.
Scores decisions, not words
The rubric scores what candidates did: what they verified, what they traded, how they handled pushback. A fluent sentence earns nothing.
AI-powered integrity monitoring.
Simulations are hard to fake; we make them harder. While candidates work, our AI watches for signs that something isn’t right: tab-switching, bot-like input, scripted responses, borrowed answers. No downloads, no lockdown browsers, no interrupting the candidate mid-flow. It runs silently in the background, so honest candidates never notice it, dishonest ones can’t get past it, and every result you see is one you can trust.
Browser and cursor signals
Tab-switching, copy-paste jumps and bot-like movement patterns, flagged in real time.
Response pattern analysis
Answers that arrive too fast, too polished or too pasted are flagged for review.
IP and device checks
Repeat attempts, shared devices and setups that don’t match the candidate are caught.
Similarity detection
Identical or near-identical responses across candidates are caught automatically.
Automation detection
Scripts, auto-fillers and AI agents posing as candidates are identified before they touch your results.
