InterviewSignal
Free · Open source · No assessment quota

When every candidate uses AI, code quality converges.

Output is no longer signal. InterviewSignal captures the full candidate work trail — every prompt, tool call, diff, and git commit — and grades how they think, not what they submit.

$pip install interviewsignal && interview install

How it works

01

HM creates an interview

Write a problem, define a rubric, set a time limit. Get back a code like INT-4829-XK. Share it with 5 candidates or 500.

02

Candidate installs and starts

One pip install. Candidate works in their own IDE with their own AI tools — Claude Code, Codex, Gemini, Cursor. No sandbox, no browser IDE.

03

Every move is captured

Every prompt, tool call, file change, and git commit is logged in a tamper-evident hash chain. The session knows who drove the thinking.

04

Candidate submits, HM triages

Auto-graded against the rubric on submit. HM reviews ranked submissions in the dashboard. 200 candidates in 15 minutes.

The signal stack

🔗

Hash-chained event log

Every prompt, tool call, and file write is chained. Any tampering breaks the chain. Gaps in the log are their own red flag.

🤖

AI-native grading

Graded on how they used AI — not whether they used it. High-leverage use scores well. Paste-and-accept scores poorly.

📂

Per-prompt git commits

Every candidate prompt silently creates a git commit. You see the code evolve step by step, not just the final diff.

🚩

9 automated flags

Too fast, too few interactions, uniform timing, hooks gap, diff mismatch, prompt ratio — all surfaced automatically.

Zero setup for candidates

pip install, one command, done. Works in Claude Code, Codex, Gemini CLI, Cursor, and Aider.

🔒

Your infra, your data

Community relay on Vercel + Supabase by default. Self-host on Fly.io for full isolation. No telemetry, no tracking.

vs Saffron and OpenRound

They validate the category. The default shouldn't be another locked-in SaaS.

Saffron (YC)OpenRoundInterviewSignal
ModelManaged SaaSManaged SaaSOpen-source infrastructure
EnvironmentBrowser IDE sandboxHosted browser + CLICandidate's own IDE
AI toolsPlatform-controlledPlatform-controlledAny — Claude Code, Codex, Gemini, Cursor
Code evolutionSession replaySubmission tracePer-prompt git commits
Tamper evidenceSession replay (vendor)Not statedHash-chained + 9 automated flags
Assessment quota5–15/mo + $49 extra5–20/mo + customUnlimited
PricingFrom $199/moFrom $149/moFree forever
Interview dataVendor serversVendor serversYour servers
Self-hostedNoNoYes
Open sourceNoNoYes
Trust modelVendor-verifiedVendor-verifiedCryptographically verifiable

Ready-to-use interview problems

Backend, MLE, frontend — real engineering problems with rubrics you can copy directly into the dashboard.

Browse problems →

Start interviewing in 2 minutes.

No platform lock-in. No setup call. No assessment quota. Free forever.

$pip install interviewsignal && interview install