How it works

One conversation.A scored breakdown that tells you what to fix.

No essays. No multi-week feedback loops. Just a live role-play, then a fast and detailed report so you know exactly which sub-skill is dragging your band down.

Coming soon to iOSFree during early access
What a session looks like

Pick. Practice. See your weak spots.

  1. 01

    Pick a case

    Browse by topic, filter by criterion, or hit Recommended. The picker biases toward your two weakest bands and topics you haven't touched.

  2. 02

    Practice live

    A 5-minute role-play with an AI patient who replies to what you actually said. Your role-card lists the tasks; the patient stays in character throughout.

  3. 03

    Get scored

    About 30 seconds after you finish, the full report lands: 0–500 overall, all 9 criteria, cue hit/miss per task, mispronounced word list, and the case you should practice next.

Under the hood

A five-stage scoring pipeline.

Each session runs through five stages — transcription, pronunciation, cue verification, criteria evaluation, and recommender. Each stage is independently testable, which lets us measure agreement against examiner calibration sets and improve over time.

  1. 01 · 01

    Speech-to-text

    High-accuracy speech transcription splits the conversation into candidate vs roleplayer turns. Word-level timing is preserved for downstream scoring.

  2. 02 · 02

    Pronunciation assessment

    A specialised pronunciation engine scores every syllable in your candidate turns. We aggregate to per-word accuracy and rank the words that hurt you most.

  3. 03 · 03

    Cue verification

    For each task on the role-card, an LLM verifier searches the transcript for evidence the cue was addressed. Returns hit / partial / miss + the quote.

  4. 04 · 04

    9-criteria evaluation

    A purpose-built rubric prompt scores each of the 9 OET criteria 1–5 with examiner-style reasoning. We aggregate to a 0–500 overall and a band letter.

  5. 05 · 05

    Next case selected

    A weighted recommender picks your next case to target your weakest two criteria + an under-practiced topic. Repeat until the bands move.

Try it the moment it ships. Free during early access.

Drop your email and we'll let you know when your TestFlight build lands.