A live first interview, the same for every founder you select
A screen-share pitch, then an AI interviewer asks your questions out loud. Only the transcript is assessed, never video, accent or polish. Text always works.
- Live AI interview First release
- Text interview First release
- 1. Pitch
- 2. Questions
AI interviewer · live captions
Slide 4 says 20 enterprise customers. How many of them pay today?
Your transcript
- 07:34Priya: Twenty brokers have signed with us.
- 07:42Priya: Twelve of those are unpaid pilots that convert in January. Eight pay annually.
Founders know what happens before the call starts
Interviews arrive in the first release. An analyst sends the invitation from the deal, with partner approval if your fund requires it. The founder starts whenever suits them, and nobody from your fund needs to join.
- 1
Branded invitation
An email from your fund with a link to your portal, valid for a period you set, with reminders before it expires.
- 2
The consent card
The purpose, the two parts and the total length, how answers are assessed, who sees the recording and for how long. A separate tick for camera, microphone and screen recording.
- 3
Device and network check
A camera preview, a microphone level, a test screen share and a connection test. A warning that a shared screen can show notifications, with advice to share one window.
- 4
Or the text interview
Always one click away, with the same questions and the same rubric, on any device.
Your interview with Northwind Seed
- One live call, 20 minutes at most. Pitch with your screen shared: up to 5 minutes. Questions from an AI interviewer: about 15 minutes.
- We assess what you say and the evidence you give, from the transcript. Never accent, fluency, appearance, the video or presentation polish.
- Only the Northwind Seed team sees the recording, kept as the fund's notice says. Share one window, not your whole screen.
Needs a desktop browser to share your screen. A phone works for the text interview.
Tick the three consents and click through. Nothing is recorded on this site.
First the founder pitches, then the AI asks your questions
Pitch: The founder shares their screen with the camera on and pitches, usually from the deck, for about 5 minutes. The AI listens. Questions: An AI interviewer asks your core questions out loud, then follow-ups drawn from the pitch, the deck and open claims, until the cap you set. Your fund sets the cap for the whole call, usually 15 to 20 minutes.
Pitch with screen share
The founder shares their screen, usually the deck, with the camera on. You set the pitch length, 5 minutes by default, with a visible timer and a gentle cue a minute before the end.
Questions asked out loud
An AI interviewer in one neutral voice, with live captions and a Repeat question button. The founder answers out loud and can keep sharing the screen to show the product.
Your core questions first
Set per interview template, in the same order for every founder, each with a priority so the most important ones always fit.
Follow-ups from the evidence
Drawn from the pitch, the deck and open claims, such as how many of the 20 customers on slide 4 pay. At most one per question, with a cap you set.
Time kept in code
The interviewer knows the time left, skips lower-priority questions to fit, closes politely a minute before the cap, and the call stops at the cap.
The same for every founder
Everyone on a template gets the same core questions, the same interviewer persona and the same time plan.
What the AI interviewer never does
- Give advice or opinions about the company.
- Comment on how the founder speaks, looks or presents.
- Ask about protected characteristics such as age, religion or family.
- Reveal a score, or say how the interview went.
- Follow instructions hidden in what the founder says or shows: their words are treated as data.
The AI reads the transcript. People watch the recording.
Delsivo assesses the substance of what was said and the evidence behind it, using the transcript only. It excludes accent, fluency, presentation polish, appearance, background, video and screen images.
| What | Means | In the assessment |
|---|---|---|
| Substance | What the founder says about the problem, customers, numbers and plan, in the pitch and the answers | Assessed against your criteria, from the transcript |
| Evidence | Whether an answer is backed by something checkable, and how it matches the deck | Assessed, with timestamped excerpts |
| Accent and fluency | How someone speaks English, pace, filler words, pauses | Excluded. Speech metrics are not computed for investors |
| Presentation polish | Confidence, delivery, slide design, how rehearsed the pitch sounds | Excluded |
| Video and screen images | How someone looks, their room, their camera, what their screen shows | Excluded. The AI never sees video or images. People watch the recording |
The default interview rubric
Funds edit the criteria and weights in each rubric version. Criteria without enough evidence are shown as “not enough evidence”, not scored low.
- Problem insight
- Solution and differentiation
- Traction and evidence behind the metrics
- Market understanding
- Go-to-market and sales motion
- Moat and defensibility
- Financials and unit economics
- Team and execution
- The ask and use of funds
Live, voice-only and text, same rubric
Before interviews ship, a paired set of video, voice-only and text interviews with the same content must score within a set tolerance of each other. It is a release gate in our evaluation suite, re-run when the model or rubric changes. We will publish how it is measured. It reduces one known risk; it doesn't replace people reviewing every assessment.
What is recorded, with whose consent, and for what
The call runs over WebRTC and is recorded on the server, so a crashed browser loses nothing already sent. Each recording needs its own consent.
| Recording | Consent | Used for |
|---|---|---|
| Camera | Its own consent tick | Your team sees the founder while they pitch and answer |
| Microphone | Its own consent tick | The transcript the AI assesses, and the audio your team plays back |
| Shared screen | Its own consent tick | The deck or product the founder showed, beside the transcript |
| Review video | Made from the three recordings above | One file with the screen large, the camera inset and both voices, with a bookmark per question |
| Transcript | Part of the interview | Timestamped and labelled by speaker. Founders can flag lines they think are wrong |
Rejoin after a drop
If the connection drops, the founder rejoins the same session within 10 minutes and continues where it stopped, in one interview record.
One retry after a failure
A session that failed for a technical reason the system recorded can be retried once. Never to improve answers.
Desktop for screen sharing
Chrome, Edge, Firefox or Safari on a computer. On a phone, the founder switches device or takes the text interview.
The recording, the transcript and an analyst who decides
Your team reads the assessment next to the moments it came from, then accepts, changes or rejects each criterion.
Player with bookmarks
A bookmark for the pitch and each question, a switch between the review video, screen only and camera only, captions and playback speed.
Cited excerpts
Every criterion cites the transcript and the exact time, such as 07:42, and jumps there. A quote that isn't in the transcript is dropped.
Analyst override
Accept, reject or change any criterion, always with a reason that stays in the record. Team ratings can stay hidden until each reviewer has rated.
Claim updates
Answers that change a deck claim become suggested updates, such as "12 of the 20 customers are unpaid pilots", for an analyst to accept.
Transcription flags
Low-confidence segments are marked for analysts, and founders can flag lines they think are wrong.
Manual path
An analyst can run a live call instead and record the notes and a consented transcript in the same place.
Built so a founder can do their best work
A first interview is often the founder's first real contact with your fund. It should feel fair.
- Start when it suits them, with no meeting to book.
- Pitch from their own deck, then answer questions with captions and a Repeat question button.
- Choose the text interview at any point before the call, with the same questions and rubric.
- Rejoin after a dropped connection and carry on.
- Read the transcript and flag lines that are wrong.
- Never see scores, ratings or internal notes.
Expiry, reminders and retention
First releaseInvitations expire after a period you set, with reminders only while the interview is outstanding. Your fund sets retention for each recording and the transcript separately, for example deleting camera and screen recordings after 90 days while keeping the transcript until the deal closes. Withdrawing consent deletes the recordings.
Questions about interviews
Try live interviews on selected deals
Pilots turn on interviews for a few deals once the first release arrives, so your team can compare them with first calls.