Four steps. One honest verdict.
Same probing depth, same evidence standard, same verdict bands a real Bar Raiser uses.
- 01
Pick the company
Amazon today; Google, McKinsey, Goldman, Meta, JP Morgan as the calibrated content for each lands. Each company has its own probe rules and rubric — not a generic interviewer.
- 02
Pick the role
Six roles (PM, PgM, AM, sales, marketing, engineer) reweight which dimensions get probed hardest. Skip if unsure — the engine picks default weights.
- 03
Talk to Maya
Voice-first. Maya probes for evidence the way a real Bar Raiser does — customer voice, hard numbers, named people, the decision rationale. No clipboard. Barge in mid-sentence; she stops.
- 04
See the verdict
Two-pass evaluator scores against the rubric, then writes a verdict, scope read, and one transcript-anchored coaching note for your weakest moment. No hallucinated quotes — every line cites your transcript.
What the report actually looks like.
Live preview — the same component your report renders, with representative sample data.
You'd be Hire.
Here's why.
Solid signal across the indicators with one gap: the customer voice in your story didn't surface until probe three. The room would lean Hire, but a real Bar Raiser would push on whether you heard the customer or constructed the rationale after the fact. Tighten that and the next loop reads as Strong Hire.
Five bands. No flattery.
Most prep tools optimise for “you did great”. We tell you the band the actual interviewer would have given. If you got a No Hire, you get the line that lost the room and the Strong-Hire rewrite — read it before your next run.
- Strong HireTop 5% — shippable answer.
- HireSolid signal, minor gaps.
- BorderlineMixed. The next probe decides.
- No HireMaterial gaps the room would flag.
- Strong No HireStop-the-room moment.
Calibrated against 16 Amazon Leadership Principles — and the equivalents at every other company.
Amazon's 16 LPs (Customer Obsession, Ownership, Bias for Action, …) each have their own indicator rubric, anti-pattern list, and probe rules — calibrated against 200+ real interviews. McKinsey's rubric is its own. Goldman's is its own. The evaluator picks the right framework per session, then scores against that company's actual bar — not a generic “leadership score”.
Models are multipliers — the rubric does the work.
Voice transcription via Deepgram. Speech synthesis via Google Chirp3-HD. Probing and scoring on Anthropic Haiku 4.5 (latency-fast for the deterministic 80%) with Sonnet 4.6 escalation for the judgment moments — creative probes, borderline calls, the end-of-session synthesis. Prompt caching cuts cost ~90% and keeps latency under two seconds. We pin model IDs and roles in the wrapper so a model upgrade is a config change, not a rewrite.
Ready for an honest verdict?
Eight minutes. No card. No email until you see your report.
Try a free Bar Raiser session →