Home · Voice capture · My Plan · Fuel
<aside> 🎙️
VoxFit is a voice-first fitness and nutrition logging app — speak your workout or meal, and AI turns it into structured, trackable data in seconds. This page is the living product record: what shipped, why, and what's next.
</aside>
| Platform | Mobile-web PWA + native Android (Google Play), one Angular/Ionic codebase via Capacitor |
|---|---|
| Core Stack | Angular 20 (signals) · Ionic 8 · Tailwind v4 · Supabase (Postgres, Auth, Edge Functions, pg_cron, pg_net, Vault) · Groq Whisper for speech-to-text |
| AI Engine | Speech-to-text: Groq whisper-large-v3-turbo, server-side, with a gym-vocabulary biasing prompt. Structuring: Google Gemini 2.5 Flash — single-shot extraction (workout parsing, meal suggestion, nutrition estimation, muscle-group classification) and a tool-calling agent loop (plan generation, weekly coach) |
| Design System | Dusk — plum-ink gradient canvas, glass surfaces with a rim-light, five role-assigned accents, Poppins 600 + JetBrains Mono |
| Status | Logging, analytics, the AI coach layer, and progress intelligence all shipped and in daily use · voice capture rebuilt on record-and-transcribe (Aug 2026) · 16 versioned migrations (0000–0015, the full schema rebuildable from the repo) plus per-user AI rate limiting · pre-launch on Google Play |
| Deep dives | PRD: AI Workout Planner & Progress/Wellness Coach · ‣ · ‣ |
| Code | github.com/amit429/VoxFit |
Most fitness logging apps ask users to do the thing that makes them not want to log: stop mid-set, open an app, tap through forms to record sets, reps, weight, or search a food database to guess at macros. That friction is why logging habits die within a couple of weeks for most people.
VoxFit's founding bet: if logging is as fast as saying it out loud, people will actually keep doing it. "Did three sets of ten twenty thirty on bench" should become structured, trackable data without the user ever touching a keyboard — and the same should be true for meals: describe what's in the pantry, or what you crave, or what you just ate, and let AI handle the structuring, the macro math, and (for meals) the recipe.
That's the product thesis every feature below builds from.
| The old way | The VoxFit way |
|---|---|
| Stop mid-set to open an app and tap through forms | Tap once, say what you did, keep training |
| Search a food database and guess at portions | Describe the meal out loud, AI estimates the macros |
| Logging habit dies within a couple of weeks | Friction low enough to become a daily habit |
<aside> 📱
About the screens on this page. These are device-framed design mockups of the Dusk system — the design source of truth, and a close match for the shipped UI, but renders rather than captures of the running app. Three details in them are deliberately not in the build, each for a reason in Future Scope below: the notification bell on Home (there is no notification centre — that slot is the profile avatar), the live transcript on the voice capture screen (no longer producible — the app records a clip and transcribes it in one call, so there is no word-by-word stream to render; the screen shows a live microphone level instead, which moves only when the mic is genuinely hearing you), and Reminders / "Export my data as CSV" in Settings (no infrastructure behind either yet).
</aside>
Tap-to-talk capture of a full workout session. The app records a bounded audio clip on the device and transcribes it in one server-side call to Groq whisper-large-v3-turbo; the transcript then goes to Gemini for structuring. Gemini 2.5 Flash parses natural, messy speech — including repeated phrases and garbled mobile speech-to-text — into structured sets, reps, weights, and cardio segments (with support for pyramid sets and multi-segment cardio). The user reviews and edits the AI's extraction before anything saves — their data, their control. Personal records are detected automatically and surfaced as badges the moment a session is logged.
Home — voice orb, streak pill, today's fuel
Voice capture — tap to record
Session result — hero stats, coach note, per-set breakdown