Your application
Product trial
To end, say bye or ask Rhea to wrap up, or press End call. Closing this tab also ends the call.
Talk to Rhea
Rhea is the AI interviewer. Ask about the role, the process or the product, get unstuck on the brief, or tell Rhea anything you want Hetvi Kidecha (HR) and the hiring team to know: a concern, feedback, a suggestion, a request. Nothing on these calls is assessed. Up to 15 minutes each.
All calls are in English. Use headphones if you can; your browser will ask for the microphone.
Before the call
Notice
To end, say bye or ask Rhea to wrap up, or press End call. Closing this tab also ends the call.
Notes passed to the hiring team
What Rhea noted for the team from your calls. The team reads every note.
Your brief
Assess the engineering quality of Smallest's products the way a customer's engineer would, then write a QA report. Pick one area and go deep. Judge the engineering, not the ML models: not whether a voice sounds natural, but whether the product does what its docs say, fails clearly and recovers.
- APIs
- Text-to-speech and speech-to-text over WebSocket and REST, against the docs: error semantics, limits, streaming.
- Atoms product
- Building an agent, call quality, turn-taking, and what a customer can see about their calls.
- Developer experience
- Docs accuracy, SDKs, time to the first working call.
- Reliability
- Reconnects, timeouts, a few requests at once.
Set up
- Sign up at app.smallest.ai with , the email you applied with. We look at usage on that account only.
- New accounts get $10 of credit. When it runs out you can ask for one extra $10 on this page.
Rules
- No load or stress testing: at most about five requests at once.
- No security scanning or penetration testing.
- Only your own account and your own data.
- A security problem goes only in the report's private security field, never in a finding or on a call. Only the person who owns security reads it.
- Two or three focused hours is enough; stop after 24 hours of work in total. We score the quality of your evidence, not the number of findings.
- Redact keys and tokens in what you paste:
sk_...is enough.
Watch out
- Renting a phone number bills $10 on the 1st of each month and can wipe your balance. You don't need one: call your agent from the browser.
- The workflow editor is blocked on the free plan. Build your agent with the single-prompt editor.
The report
Scope and why, your method, your findings, what is done well, your top three fixes with why, and the time you spent. Each finding: a title, what you expected and what you observed, the evidence (commands, requests, request IDs, times with your time zone, numbers from several runs), what a customer would see against an SLO, and a severity.
A well-written finding, from another product
It shows the shape, not an answer: the product here is not Smallest's.
- Title
- 429 responses carry no Retry-After header
- Expected
- The rate-limits page says every 429 carries Retry-After in seconds.
- Observed
- 19 of 19 responses with status 429 had no Retry-After header and the body {"error":"too many requests"}.
- Evidence
for i in $(seq 1 30); do curl -s -o /dev/null -w '%{http_code} %{time_total}\n' -X POST https://api.example.com/v1/geocode -d @addr.json; doneRun 2026-10-28 14:05 to 14:07 IST: requests 1 to 11 answered 200 in 180 to 240 ms, 12 to 30 answered 429. Request IDs req_8f3a21 and req_8f3a22 from the X-Request-Id header.- Customer impact
- A client that honours Retry-After retries at once, so a burst becomes a retry storm. Against a 99.9% success SLO, 63% of a 30-request burst failed; a customer importing addresses sees most of the import fail.
- Severity
- major
Smallest may use the product feedback in your report to improve its products.
Your Smallest account
Your QA report
Saves as you type. Rhea reads it before the technical call, which starts from it. Starting the technical call submits it and locks it.