Topics picked for you
Speaking topics picked for you, so you practise thinking on your feet — not reciting a script
- Impromptu topics
- Up to 2 min
Our productPublic beta
AI × Speech & Vision Coaching
VakSutra AI
An AI speaking coach that watches and listens like a human coach would — then shows you, in your own words, exactly what to fix.
vaksutra-ai.me
Product walkthrough · sample session. Screens recreated from the live app; scores shown are sample values, not real user data.
- 01Get a speaking topic — e.g. “Describe a time you changed your mind about something.”
- 02Record up to two minutes on camera
- 03Get an overall score with eye contact, filler words and pace
- 04See the proof moment: your original sentence next to a stronger rewrite
- Per practice answer
- 2 min
- Areas scored
- 4
- During public beta
- Free
The problem
What needed solving.
Most people go blank when they have to speak without preparation — in an interview, a meeting, a viva or a presentation Q&A. Practice is the only real fix, but a human coach is expensive and rarely available the night before it matters.
Practising alone doesn't work well either. You can't watch your own eye contact, count your “ums” and judge your pace while you're trying to think — and most AI tools reply with generic advice like “be more confident”, which gives you nothing to act on.
BeforeExpensive coachingPractising blindGeneric AI tips
What was engineered
Scoring in code, coaching on top.
One score, four areas
Practise in the browser on camera, for up to two minutes per answer
- Fluency
- Language
- Voice
- Presence
Your own sentences, rewritten
One overall score across four areas: fluency, language, voice and presence
- Grounded rewrite
- Consistent scoring
Feedback tied to exact moments
Your main weak spot named clearly, so you know what to fix first
- Timestamps
- Playback
Architecture
How it fits together.
Simplified for this page. Highlighted parts carry the core logic.
Input01
Camera
face and posture
Microphone
voice
Processing02
MediaPipe
478 face landmarks
Whisper
transcription
Rust DSP
acoustic signal processing
Output03
Scoring in code
4 pillars
Grounded rewrite
coaching from your transcript
Tech stack
- Next.js 15 + TypeScript
- MediaPipe (WebAssembly)
- FastAPI + Python
- Rust (PyO3)
- Whisper large-v3 via Groq
- Mistral + Groq LLMs
- Supabase (Auth + Postgres + Storage)
- Vercel + Render
Like what you see?
Let's build something together.
Tell us what you want to build. We reply within 24 hours with questions or a first outline of scope and approach.