The AI takes the keyboard

An AI interviewer thatwatches

SViam holds a real spoken interview, watches every keystroke, and pushes back the moment you go wrong. It rewrites your lines, runs the code, and signs a hire decision with timestamped receipts.

  1. 01watchesevery keystroke
  2. 02interruptsmid-keystroke
  3. 03editsyour code, live
~30-minute screen/One link · one PIN · zero engineer hours/Tamper-evident, receipt-backed reports
solution.pyPython 3.12
AI interrupts·AI rewrites your code·tamper-proof receipts

A real interview. The AI interrupts, edits the candidate’s code, and cites every moment.

What it does

It watches how they think, not just what they submit.

Candidates work in a full repo with tests, lint, and a running dev server, the way they actually ship, not a whiteboard fantasy.

src/
api/route.ts
lib/rate-limit.ts
tests/limit.test.ts
package.json
dev server ready · localhost:3000
Auto-generated scorecard

A defensible decision, ready before they close the tab.

Every interview ends with a structured, rubric-backed report. No note-taking, no recency bias, no "I think they were fine."

  • Reasoned out loud before touching the keyboard
  • Recovered cleanly after the curveball follow-up
  • Spent 40% of time on tests, unusually high

Candidate

Senior Frontend · Loop 2

8.6/10
Advance

Clears the bar on signal and integrity. Move to onsite.

Reasoned out loud before touching the keyboard

Recovered cleanly after the curveball follow-up

Spent 40% of time on tests, unusually high

Rubric breakdown

Problem decomposition
92
Code correctness
88
Communication under pressure
95
Edge-case handling
71
Iteration speed
84

Summary

Strong communicator who narrates trade-offs clearly. Slight gap on edge cases under time pressure, worth a targeted follow-up in the onsite.

~30 minA full technical screen
One linkPlus a PIN, no scheduling
ReceiptsEvery claim timestamped
0Engineer hours to run it
Built with 16 hiring managers · Backed by PESP

Stop guessing. Start watching them build.

Run your next technical screen with SViam and get a defensible, rubric-backed decision before the candidate logs off.