</speechmax>

────────────────────────────────────────
[TEAM PROJECT][WEB APP][HACKATHON WINNER]
reacttypescriptmediapipeweb-speech-apiweb-audio-apigemini-2.5-flashsupabaseframer-motionzustandtailwind

</the problem>

────────────────────────────────────────

Speech therapy is boring and people drop off. 77% of people have public speaking anxiety but professional coaching costs $200/hour. Only 8% ever seek help. Existing tools are clinical and feel like homework, not something people stick with.

</my approach>

────────────────────────────────────────

Built a Duolingo-style progression system with real clinical speech therapy methods. Real-time webcam analysis using MediaPipe for eye contact and posture, Web Speech API for transcription, and Web Audio API for pitch detection. All processing runs client-side for privacy. Designed five targeted mini-games that each isolate a specific weakness.

</key features>

────────────────────────────────────────

Real-time webcam analysis

468 facial landmarks for eye contact tracking, 33 body keypoints for posture and fidget detection, all running at 30+ FPS client-side.

Five gamified training modes

Filler Ninja, Eye Lock, Pace Racer, Pitch Surfer, and Stage Presence. Each targets a specific weakness with progressive difficulty.

AI coaching with Mike

Gemini 2.5 Flash powered coach that sees your scores and game history, giving short actionable advice.

Radar scan scoring

30-second assessment across five axes calibrated against Toastmasters evaluation criteria.

</what i learned>

────────────────────────────────────────
>

Balancing fun game feel with genuine therapeutic value requires grounding gamification in real clinical research (deliberate practice, Ericsson 2008).

>

Building under 48-hour pressure forces ruthless prioritization. We cut features that felt important but weren't essential to the core loop.

>

Client-side ML inference eliminates privacy concerns entirely, which matters deeply for something as personal as speech coaching.

[← back to projects]