Human resources
Interviews that hold the whole conversation
Video and coding interviews that remember what the candidate said forty minutes ago, follow up on it, and score against the same bar every time.
The problem
Screening doesn't scale and doesn't stay consistent
A first-round screen is forty minutes of a senior person's time, and the tenth interview of the week is not scored like the first. Different interviewers probe different things, so two candidates for the same role are evaluated against two different bars.
AI screening was meant to solve this, and mostly produces something worse: a rigid question list that can't follow up, can't reference what the candidate said earlier, and evaluates a transcript rather than a conversation.
AI interviewer
Long-haul interviews, video and coding
A forty-minute interview is one conversation, not forty separate minutes. Holding it as a single session is what makes the later questions worth asking.
Context across the full session
References what was said at minute five when probing at minute forty. Follow-up questions build on the candidate's actual answers instead of moving down a list.
Video interviews
Structured conversation with real follow-up. The interviewer adapts depth based on what the candidate demonstrates, the way a human would.
Coding interviews
Watches the approach, not just the final submission. Asks about tradeoffs the candidate made, grounded in the code they actually wrote in the session.
Consistent scoring
Every candidate is evaluated against the same rubric, held in memory. The tenth interview of the week is scored like the first.
Why memory is the hard part
A stateless interviewer isn't an interviewer
An interview is one long conversation where the value is entirely in the follow-up. Without session memory the model can only ask what's on the list — and a candidate who gives a revealing answer at minute ten gets the same next question as one who didn't.
Session memory is what turns question-asking into interviewing. It's also why most AI screening tools feel like a form with a voice.
Fairness and record
Auditable by construction
Every interview leaves a retrievable record — what was asked, what was answered, and what the score was based on. When a hiring decision is questioned, the reasoning is available rather than reconstructed.
The interviewer produces the evidence; a person still makes the decision. That separation is deliberate, and it is what the record is for.
FAQ
How long can an interview run?
Session memory has no practical turn limit — state lives outside the context window, so length is not constrained by the model.
Does it replace human interviewers?
It replaces the first-round screen. Final rounds stay human, with the screening record available as context.
Where is candidate data stored?
On your infrastructure. Self-hosted means candidate recordings and transcripts never leave your network.