Spoken English practice app design quiz

01

MCP boundary

What was the main weakness in the first TalkLooper architecture?

A. It used Supabase instead of storing all state in the iOS app
B. It let Claude be the only audio path, so cleaned transcripts could hide the learning signal
C. It used SwiftUI views instead of a UIKit coordinator, even though UIKit would not solve transcript cleanup or speech evidence
D. It had no way to store user profiles, practice sessions, or observations

02

Raw speech signal

A. Because language learning often depends on fillers, pauses, pronunciation, timing, and self-repair that may not survive transcript cleanup
B. Because transcripts are never useful for coaching, progress views, debrief summaries, search, scenario prompts, or any future practice loop
C. Because iOS apps cannot receive text from a backend unless they also upload audio
D. Because MCP requires every tool call to include an audio file and a waveform image

03

Speech event pipeline

A. To replace every user-facing observation with raw JSON rows, timestamps, confidence values, and chunk ids in the app UI
B. To store typed evidence such as fillers, pauses, repetitions, confidence, and timing before creating coaching observations
C. To force the mobile client to run every analysis model locally before upload
D. To remove the need for user consent because events are smaller than audio

04

Consent and control

05

Backend analysis

A. Because analysis can take longer than the user-facing session end, and the app can show results when workers finish
B. Because backend workers cannot write to Supabase after the user leaves the practice screen, locks the phone, or closes the app
C. Because Realtime updates only work while audio is actively recording
D. Because a practice app should never show a debrief

06

Data model

A. Because observations should stay the durable user-facing coaching unit while lower-level evidence can have different retention and debugging rules
B. Because Postgres tables can only store one type of text column per user, session, practice scenario, and observation row
C. Because Supabase RLS does not work for observation rows
D. Because audio chunks must be displayed directly on the home screen

07

Feedback design

A. You are bad at speaking clearly and should stop using fillers
B. In this answer, you opened with um and paused before the main point. Try: The main tradeoff is latency versus privacy
C. Your speech score is lower than most users in your cohort this week
D. Here is a complete dashboard of every weakness detected in the session, sorted by severity, frequency, confidence, and comparison to other users

08

Mobile reliability

A. Keep local chunks for retry with sequence numbers and idempotency keys, then let the user discard if they want
B. Silently delete the session because partial uploads are impossible to repair
C. Create duplicate observations every time the same chunk is retried
D. Block the user inside the recording screen until every upload, transcript job, speech event extraction, and model call completes