Skip to content
GA George Andary
Menu
IN PLAY MATCH REPORT SEP 2026

Babel Fish

French-practice prototype; device audio unverified. Five-card sessions, manual phrase conversion, and saved cards passed review.

Ask about this build
Babel Fish prototype start screen with French sentence patterns and practice navigation

The read

Recognizing a phrase is only one part of practice. A learner also needs a simple way to repeat its structure, save useful examples, and return to them later.

Game plan

  • Organize static French examples into reusable sentence patterns.
  • Connect a five-card session to repetition, self-ratings, and saved cards.
  • Let a learner supply a translation, phonetics, and context to turn an original draft into a practice card.
  • Keep saved cards and activity in the browser without requiring an account.

Final score

A connected practice flow from sentence patterns to a personal library, with explicit boundaries around feedback and local storage.

  • A correction walkthrough completed five cards, five repeats, and five self-rating changes without recorded session errors.
  • Saved cards, manual draft conversion, and library and activity persistence were exercised through browser reload.
  • Structured lesson-data validation and HTTP checks passed; independent Claude review passed on September 24, 2026.
  • Audible normal and slow French playback still requires a real-device check.

The idea

Babel Fish explores French practice through reusable sentence structures: “I am,” “You are,” and “I want.” A learner can practice static examples, save a useful card, or manually turn an original phrase into something to revisit.

My contribution is product direction and scope. AI assisted implementation and verification of the JavaScript interface, structured lesson data, and browser persistence.

The working flow

  1. Start a five-card session from Patterns.
  2. Repeat and self-rate the cards, save an example, and finish the session.
  3. Create a draft and manually supply its French translation, phonetics, and context.
  4. Open the converted card in the saved library or scripted Voice Coach.
  5. Reload the browser and return to saved cards and logged activity.

The active review queue is temporary. Saved cards and logged activity persist in that browser; clearing browser data removes them.

What we verified

Data validation, HTTP routes, the five-card completion flow, manual conversion, and save/reload behavior were checked. The final correction capture showed five practiced cards, five repeats, and five self-rating changes. Those counts demonstrate the interface working; they do not measure learning gains.

Independent Claude review passed on September 24, 2026. Audible playback remains unverified, so the full demo milestone is still open. The screenshot above shows the local prototype with sample content.

Honest limits

Voice Coach offers scripted guidance and self-ratings. It does not record, transcribe, or assess speech. Translations are manually supplied, and topic search uses local examples. There is no account sync or evidence of improved fluency.

Browser speech controls are implemented, but a successful click or an available voice does not establish that a person heard the output.

Next milestone

Verify normal and slow French playback on the intended device, then gather learner feedback on whether the short practice flow is useful.

Guided walkthrough of the verified core flow available on request.

NEXT FIXTURE Rugby Reads →