For Engineering & Recruiting Teams

Hire developers whoactually use AI well.

"AI-fluent" is the new "team player": everyone writes it, nobody can prove it. And LeetCode was never going to. So we built CrowdVibe. A sandboxed IDE with an AI assistant on the side, scenarios tuned to the role, and a full report plus session replay at the end. Real signal, not vibes.

No setup · Sandbox provisioned per candidate
5Scoring Dimensions
6Report Tabs
100%Replayable
CrowdVibe Assessment, Scenario 1 of 3
12:34
Fix the Auth|Find and fix security issues in the authentication module
auth.ts middleware.tsutils.ts
1import jwt from 'jsonwebtoken'
2import {'{'} compare {'}'} from 'bcrypt'
3
4const SECRET = 'secret123' // ← hardcoded
5
6export const verifyToken = (token: string) => {'{'}
7 const decoded = jwt.verify(token, SECRET)
8 return decoded // no algorithm check
9{'}'}
10
11export const login = async (email, password) => {'{'}
12 const user = await findUser(email)
13 if (!user) return {'{'} error: `No user: ${'{'}email{'}'}` {'}'}|
14 const valid = await compare(password, user.hash)
AI Assistant
Claude Sonnet
You · 2 min ago
Review this JWT implementation for security vulnerabilities. Focus on authentication bypass risks.
AI · just now
I found 3 critical issues:

1. Hardcoded secret, SECRET is a string literal, trivially discoverable
2. No algorithm constraint, missing algorithms option enables "none" algorithm attack
3. Error message leaks user existence, line 13 reveals whether emails are registered
Ask about the timing attack on line 14...
Connected TypeScript · Node.js 5 prompts · 847 tokens 3 pass 5 fail

Every candidate "uses AI." Almost none use it well.

The day-to-day of engineering changed. Hiring didn't. Most coding tests still ban AI or score against rubrics from 2018, so you end up hiring for puzzle-solving instead of the skill that drives output today: working effectively with an AI partner on real tasks under real constraints.

Traditional Coding Tests
  • Algorithm puzzles unrelated to the role
  • AI tools banned (yet used in real work)
  • Pass/fail grade with no signal beneath
  • No visibility into how the candidate thought
CrowdVibe Assessments
  • Real engineering tasks tuned to the role level
  • AI assistant built in, measured, not banned
  • 5 weighted dimensions with an AI-written summary
  • Full session replay with integrity flags

From invite to hire decision in one afternoon

Four steps. No installs, no proctor calls, no synchronous interview block. Your team configures once, then reviews structured signal per candidate.

01

Configure & Invite

From your company dashboard, pick the role level (junior to staff), stack (TypeScript/Node, React, Python/FastAPI), and scenario mix. Generate an invitation link and send it, or attach it to a job listing.

02

Sandbox Provisions

Candidate clicks the link, lands in a pre-flight lobby, and a fresh isolated sandbox spins up automatically. No installs, no setup. Same environment for every candidate: same Claude model, same system prompt, same tooling.

03

Candidate Builds with AI

Three real-world scenarios in a Monaco IDE alongside an embedded AI assistant. Telemetry captures every keystroke, AI prompt, paste, test run, and focus event in real time, all replayable later.

04

Review & Decide

Get a six-tab report covering Performance, Behavior, Provenance, Grading, Integrity, and Methodology, plus a complete time-synced session replay. Share with the team, export to PDF, or compare candidates side by side.

Run scenarios in the stack your team actually ships in

Every language runs in the same isolated sandbox with the same AI assistant, telemetry, and scoring rubric. Pick the one that matches the role.

TypeScript
JavaScript
Python
Java
Go
Rust
C#
Ruby
PHP
More coming soon

Need a language not listed? Ask us, most stacks can be enabled within a sprint.

Not just whether it ran. How they got there.

Five weighted dimensions surface the signal that matters in AI-era engineering. Every dimension is graded automatically and explained in plain English.

AI Leverage 25%

Does the candidate use AI as a force multiplier or a crutch? A secondary AI model analyses prompt specificity, problem decomposition, and whether they reach for AI on architecture and edge cases, not just rote code generation.

Strong"Review this JWT implementation for timing attack vulnerabilities and suggest a constant-time comparison"
Weak"Fix the code"

Critical Review 20%

Does the candidate catch AI mistakes, or paste them straight into prod? Each suggestion they accept, edit, or reject is logged. Thoughtful evaluation scores; blind acceptance does not.

Accepted with edits+4
Caught incorrect suggestion+6
Modified AI output+3
Pasted without review-2

Task Completion 25%

Does the code actually work? Automated test suites grade against acceptance criteria, edge cases, and error-handling paths, same rubric for every candidate.

JWT validation secured
Input sanitised
Timing attack mitigated

Code Quality 20%

ESLint + Semgrep static analysis flags input validation, error handling, security anti-patterns, and structural smells. Surfaces candidates who ship sloppy AI output unchecked.

No security anti-patterns
Proper error handling
Missing input validation (1)
Clean code structure

Speed & Efficiency 10%

Time-to-completion normalised by difficulty, plus a prompt-efficiency ratio measuring productive iteration vs. aimless looping. Rewards intentional work, not raw speed or prompt volume.

Screen on your stack, your work.

Stock scenarios are a starting point, not a ceiling. Generate AI-built coding scenarios from a job description, an existing repo, or a freeform brief, then refine them in a guided builder until they reflect the work your team actually ships.

Generate from a Job Description

Paste a JD or pick an existing job listing, CrowdVibe drafts a scenario tuned to its skills, stack, and seniority. Review the brief, starter files, and test suite before publishing.

Choose the Scenario Type

Code-write, debug, code review, system design, stakeholder comms, or git workflow. Mix scenario types in a single assessment to test the dimensions of the role that matter to you.

Calibrate the Difficulty

Junior, mid, senior, or staff, difficulty calibrates the brief depth, edge cases, and grading rubric automatically. Same scenario, four bars: pick what matches the role.

Bring Your Own Starter Files

Upload starter code and a test suite, or let CrowdVibe scaffold both for you. The candidate's sandbox boots with exactly your repo layout, no synthetic playgrounds.

Add Follow-Up Scenarios

Chain scenarios that react to the candidate's submitted code, a stakeholder-comms exercise that asks them to explain their changes, or a git-workflow follow-up that ships the diff. Real engineering rarely ends at "tests pass."

Test Drive Before You Send

Run any scenario yourself in the same sandbox the candidate sees. Validate the brief, check timing, and review the report it produces before exposing a single applicant to it.

Reusable Scenario Library

Every published scenario lives in a private library shared across your team. Tag, filter by difficulty or type, and reuse across job listings, so each role keeps building on consistent signal.

Identical Grading, Custom Content

Custom scenarios still grade on the same five weighted dimensions. Your scoring stays comparable across candidates and across roles, even when the brief is unique to your company.

A signal your team can actually defend

The moment a candidate submits, your team gets a structured report with an AI-written summary, sub-score breakdown, prompt analysis, integrity flags, and a complete time-synced session replay.

CrowdVibe Results, Assessment Complete
CrowdVibe Assessment

Candidate Report, Sarah K.

AI-augmented developer assessment, Web Development (Full-Stack, Senior)

Submitted: 7/13/2026, 6:40 PM · Duration: 42 min

Overview
Prompt Analysis
Scenarios 3
Improvement Areas
Share & Badge
CROWDVIBE80out of 100
Proficient, Verified

Top 22% of Full-Stack developers

AI Assessment Summary

Strong AI collaboration with room for edge case mastery

Sarah's prompts are specific and well-structured, she consistently breaks complex problems into targeted AI queries. She caught 2 of 3 AI mistakes in the refactoring scenario. Main growth area: edge case coverage in Scenario 2 and a missed timing vulnerability in Scenario 1.

AI LEVERAGE
82/100
COMPLETION
85/100
CODE QUALITY
78/100
REVIEW
76/100
SPEED
80/100

Prompt Strategy Analysis

How effectively the candidate communicated with the AI assistant

View full analysis
Targeted Problem Decomposition
Strength

Broke Scenario 1 into 4 focused AI queries instead of asking for a complete solution, leading to higher defect detection and more deliberate fixes.

Edge Case Prompting
Improve

In Scenario 2, built the happy path but didn't prompt AI about error states, partial failures, or rate-limiting edge cases. A "what could go wrong?" prompt earlier would have caught these.

Score Tiers

Where this candidate falls in the CrowdVibe framework

Expert
90-100, Strategic AI partner. Exceptional quality and judgement.
YOU
Proficient
70-89, Strong AI usage with good judgement. Crowdcruit Verified.
Competent
50-69, Adequate AI usage with room to grow.
Developing
0-49, Building skills. Likely below the bar for most senior roles.

Defensible signal. Fair to candidates.

Designed with engineering managers and recruiters who screen technical candidates every week. No gimmicks, no leaked answer keys.

Identical Conditions

Every candidate gets the same Claude model, same system prompt, same scenario set tuned to the role level. Scores compare cleanly across your funnel.

Integrity Built In

Focus tracking, idle classification, paste-source detection, and hallucination flags surface anomalies. You see how a candidate worked, not just what they submitted.

Full Session Replay

Scrub through a candidate's session like a video, code edits, AI conversations, test runs, and focus events all time-synced. Skip to interesting moments via auto-generated markers.

AI-Analysed Reports

A secondary AI model writes the executive summary, classifies the candidate's persona, and explains sub-scores in plain English, so non-technical stakeholders can follow along.

Sandbox Per Candidate
6-Tab Hiring Report
PDF Export & Share
Integrity Flags

Frequently asked questions

Everything hiring teams need to know about CrowdVibe.

CrowdVibe measures how candidates actually collaborate with AI on real engineering work. We score five weighted dimensions: AI Leverage (25%), Task Completion (25%), Critical Review (20%), Code Quality (20%), and Speed & Efficiency (10%). The result is a defensible signal on AI-era productivity, not just whether the code compiled.
From your company dashboard, pick the role level and stack, generate an invitation link, and send it. Candidates click in, the platform provisions a fresh sandboxed environment, and they begin. No setup, install, or screen-share required on your side.
A short pre-flight lobby while their sandbox is provisioned, then three real-world scenarios in a Monaco-based IDE with an embedded Claude AI assistant. They write code, run tests, and chat with the AI exactly the way modern developers work. Total time: ~45 minutes.
A six-tab report, Performance, Behavior, Provenance, Grading, Integrity, and Methodology, with an AI-written executive summary, persona classification, sub-score breakdown, percentile rank, integrity flags, and a complete time-synced session replay of code, AI conversations, and test runs. Exportable to PDF and shareable internally.
Every candidate gets the same Claude model, identical system prompt, and a scenario configuration matched to the role level. Grading is automated and AI-analysed against the same rubric. Conditions are identical, so scores are directly comparable across your pipeline.
The platform tracks focus, idle classification, paste sources, keystroke bursts, and AI interaction patterns end-to-end. Integrity flags surface anomalies, large external pastes, off-task time, hallucinated dependencies, so you can review the full context before making a call.
Web development today: TypeScript / Node.js / React and Python / FastAPI, with scenario difficulty scaling across junior, mid, senior, and staff levels. Scenario types span code-write, debug, code review, system design, stakeholder comms, and git workflow. More stacks are on the roadmap.

Hire on signal, not on vibes.

Send your first CrowdVibe assessment in under five minutes. Get a multi-dimension report and a full session replay back, ready for your next hiring loop.