Research engineer. I write the protocol first, then build the thing that runs it.
Currently at Kalp Intelligence in Bengaluru, shipping production apps used by around 9,000 people.
CogniFlow reads cognitive load off a plain webcam: pupil dynamics, blink behaviour, gaze stability, facial tension. Everything runs on-device and no video leaves the machine.
The study protocol is written and pre-registered. It has not been run yet, which is currently the interesting problem.
I wrote about deleting its main metric here: I built a webcam focus tracker, then deleted the focus score
| Project | What it does | Built with |
|---|---|---|
| Katha | Turns dense technical material into ordered read-a-little, build-a-little steps. Pulls readable content out of any URL or YouTube link, Readability with a Playwright fallback for JS-heavy pages. | React, Express, Playwright |
| BahniMitra | AI learning-path generator. Curated video roadmaps with progress tracking and shareable paths. Live | React, TypeScript, Appwrite |
| SkyGuard | Air-quality monitoring for Delhi. 48-hour ML forecasting, 3D pollution visualisation, policy advisor. | Python, Prophet, Streamlit |
| Convex | Event discovery app. Geospatial search, dynamic category theming, admin moderation workflow. | React Native, Express, MongoDB |
TypeScript, Python, React, React Native, Tauri, Express, MediaPipe.
suryanshpatwa.in · blogs.suryanshpatwa.in · @suryansh_gyaani