◆AI-powered technical screening — now in private beta

Hiring engineers,
without the guesswork.

Anvil runs the entire screening loop — invite, proctor, score, and rank — so your team interviews the right three candidates instead of triaging fifty. The AI does the reading. You make the call.

Book a demo ▶Watch 2-min product tour

✓  14-day free pilot✓  No credit card✓  SOC 2 Type II↳ Live in 12 minutes

anvil.io/acme/assessments/backend-engineer

Acme /Assessments /Backend Engineer

Search candidates

Backend Engineer — Acme

84 invites · 62 completed · sorted by total score

All · 62Pass · 19Flagged · 7

#	Candidate	Total ↓	Risk	Status
01	RO Rita Okafor r.okafor@…	94	low · 4	submit
02	AM Anika Mehta anika@…	89	low · 8	submit
03	ML Mei Liu mei@…	76	HIGH · 82	review
04	JP Jordan Park j.park@…	71	low · 11	submit
05	DK Devon Kim devon@…	58	med · 24	submit

AI summary · Rita

conf. 92%

Solved the graph problem with BFS · O(V+E) — the optimal approach. Clean, idiomatic Go.

verdict94/100

advance15 min saved

Engineering teams hire on Anvil

◆ Modulr▲ Northwind◐ Verge.dev✦ Quanta⬢ Hexbase⬡ Foundry

01 / The problem

Every senior engineer interview slot is precious. Don't spend it discovering someone can't write a for-loop.

01The take-home is dead

Candidates use ChatGPT. Recruiters can't tell. The signal is gone, and the time-to-hire is six weeks.

02Whiteboards don't scale

You can't run two-hour onsites for fifty applicants. Engineering managers burn out before round three.

03The legacy tools are noise

HackerRank reports a score. Anvil tells you why — with code samples, behavioral evidence, and a one-line verdict.

02 / The product

One surface for the whole screening loop. Designed for the speed of how engineers actually work.

Six tools, one platform. The candidate IDE, the recruiter rankings table, the AI scorecard, the EM plagiarism diff — built on the same primitives so context flows freely between them.

AI scorecard

A verdict in twelve seconds.

Every submission lands with a one-paragraph plain-English summary — what the candidate built, where they got stuck, and whether they're worth your next hour.

AI assessment

confidence 92%

Rita solved Q2 with BFS · O(V+E) — the optimal approach. Q3 passes 18/20 cases, fails on disconnected graphs. Strong Go fundamentals.

Recommend advancing to system design.

Verdict

advance

Score

94/100

Live monitoring

See twelve candidates at once — without watching them.

LIVE · 38m

Rita O.Q2 · 32%

LIVE · 41m

Anika M.Q3 · 64%

FLAG · paste

Mei L.Q1 · 12%

Candidate IDE

Pyodide, gcc, Node — all in the browser.

readme.mdsolution.gotests.go
1func shortest(g map[int][]int) int {
2  q := []int{}{
3  seen := map[int]int{}0: 0}
4  for len(q) > 0 {
5    n := q[0]; q = q[1:]

Rankings

Sort 200 candidates in a breath.

RORita O.

AMAnika M.

MLMei L.

JPJordan P.

DKDevon K.

Plagiarism diff

Catch shared answers — across cohorts, across years.

Mei Liu · 2026def bfs(g, s):
  q = [s]; seen
  while q:

87%

A. Patel · 2024def bfs(g, s):
  q = [s]; seen
  while q:

match · 87%4 of 7 lines identical · including variable names

ATS sync

Pipes into Greenhouse, Lever, and Ashby.

GGreenhousesynced14 candidates

LLeversynced6 candidates

AAshbyconnect—

03 / How it works

From requisition to short-list in under a week.

The full candidate journey, instrumented end to end. Each step persists evidence so the next one starts with context, not a blank page.

Step 01 · 2 minutes

Invite — single-use link, zero account.

Pick a template, paste a list of emails, hit send. Candidates click through to a token-authed page; no signup, no password reset support tickets.

Acme Hiring · to candidate@…

Your assessment for Backend Engineer

90 minutes · 3 problems · webcam proctored

Begin assessment →

anvil.io/a/a8f3-9c2e-bb11

Step 02 · 60–90 minutes

Solve — proctored browser IDE.

Three problems in any of seven languages. Webcam baseline, tab-switch flags, and paste detection — without the hostile theatre of locking the browser.

⏱ 47:12
1def shortest(graph, start):
2  queue = [start]
3  seen = {start: 0}
4  while queue:
5    node = queue.pop(0)
6    for n in graph[node]:

Step 03 · 12 seconds

Decide — read the AI verdict, advance.

Candidates land on a ranked board, each with a one-paragraph summary, risk flag, and a one-click "advance to next round" that pushes back to your ATS.

RORita O.clean94

AMAnika M.clean89

MLMei L.HIGH76

38min

Median time saved per candidate review.

vs. recruiter screen baseline

4.2×

More candidates screened per week, same team.

Q1 2026 cohort

73%

Of EMs say first-round interviews became “useful, not redundant.”

post-pilot survey · n = 47

0.92

Correlation between Anvil verdict and on-the-job 90-day performance.

internal · 6-mo follow-up

04 / Built for

Three roles. One source of truth.

Recruiters move volume. Engineering managers protect quality. Candidates want a fair shot. Anvil is the only surface where all three meet on the same evidence.

For recruiters

Triage fifty candidates before lunch.

Rankings table with AI verdicts in one click. No more reading 90-minute recordings.
One-click push to Greenhouse, Lever, or Ashby. Notes, score, and proctor flags ride along.
Templates per role: backend, frontend, full-stack, data — ready in 3 minutes, not 3 days.

For engineering managers

Stop interviewing engineers who can't code.

Build assessments from your own codebase. Real problems, not LeetCode.
Plagiarism diff catches shared answers across candidates, cohorts, and years — even with renamed variables.
Replay the full session — keystrokes, tab events, paste origin — when you need to be sure.

05 / In the wild

“We cut our screening loop from six weeks to nine days. The AI verdict isn't perfect, but it's right often enough that I trust it for the first sort — and the proctor evidence has settled three plagiarism cases this quarter alone.”

Priya Sharma

Head of Engineering · Modulr

−72% time

Time-to-shortlist

3.8× volume

Candidates per recruiter / week

+18pp NPS

Candidate satisfaction (post-test survey)

06 / Security & trust

Designed for legal review. Built for engineering review.

Anvil treats candidate footage and code as you'd treat a patient record — encrypted in transit and at rest, region-pinned, purged on a schedule you control, audited end-to-end.

SOC 2 Type IIaudited annually

GDPR & CCPAEU residency · DPA on file

Regional pinningUS · EU · APAC

30-day purgeconfigurable retention

Bias reviewEEOC reports per role

SSO & SCIMOkta · Google · Entra

07 / Pricing

Pay per role, not per seat.

Every recruiter and EM on your team gets unlimited access. You pay for the number of open roles you're screening into — so the price scales with hiring, not headcount.

Starter

For small teams

$0/ first role · 30 days

1 active role · 50 invites
AI scorecard + plagiarism check
3 question templates
Email support

Start free pilot

Growthmost teams

For hiring engineering teams

$1,200/ role / month

Unlimited recruiter & EM seats
Custom assessments + own codebase
Greenhouse · Lever · Ashby sync
Analytics & bias review
Slack-based support

Book a demo

Enterprise

For 200+ engineers

Talk to us

Volume pricing, annual contract
SSO · SCIM · audit log export
Dedicated region (EU / APAC)
Solutions engineer on call

Contact sales

What counts as a “role”?+

A role is one open requisition you're actively screening into. You can swap which role is active at any time — invites issued under a closed role stay readable but generate no new evaluations.

How accurate is the AI scorecard?+

In our six-month follow-up study, Anvil's advance / reject verdict correlated with 90-day on-the-job performance at 0.92. That said, it's a recommendation, not a decision — every verdict ships with the code samples and behavioral evidence behind it, so a human can override in one click.

What about candidate experience?+

Candidates get a single-use link, no signup, a familiar VS Code-like IDE, and a clear briefing about what's being recorded. We don't lock the browser or require any extensions. Post-test NPS averaged +18 points over the legacy tool we replaced.

Will it work with Greenhouse / Lever / Ashby?+

Yes — native integrations for all three, plus a generic webhook + API for everything else. Notes, score, and proctor flags push back automatically. Single-direction or two-way sync are both supported.

Can I build assessments from our real codebase?+

Yes. Engineering managers can paste a sanitized snippet or upload a small repo, mark which functions are "the problem", and Anvil scaffolds tests and a rubric from that. Most teams write their first custom assessment in under thirty minutes.

◆ Ship the right hire faster

Twelve minutes to set up.
Six weeks back per quarter.

Bring a role you're hiring for. We'll wire up the assessment, send three test invites, and you'll see the first AI verdict before the meeting ends.

Book a 20-min demo Talk to sales

No credit card · 14-day pilot · cancel anytime