Photograph a medicine strip and MedLens reads it, matches it to one of 254,000 Indian medicines, lists brands with the same salts for less and checks every pair against 160,000 known interactions. A 0.3M-parameter retriever does the matching in 96 ms.
Ask questions about a video on an ordinary laptop. Most frames of a clip repeat each other, like this 1878 gallop. CLIP embeddings and k-means keep 10 of 62 frames, BLIP captions them, FAISS finds the right moments and Flan-T5 answers only from what was seen. About five times faster than reading every second.
A menu-bar app in Go that signs you back in to a campus captive portal within a second or two of waking. The password stays in the OS keychain, the portal’s certificate is pinned, and it never retries a wrong password. Windows, macOS and Linux.
Detection, segmentation, pose and tracking with YOLO in one Streamlit app: images, video files, webcam, RTSP and YouTube streams, with per-frame and running counts. The boxes on this photograph are its own output.
A webcam workout tracker. MediaPipe Pose finds 33 points on the body, the elbow angle decides when a bicep curl counts, and the count and feedback update live on screen.
Gesture Cursor (2025) reads 21 points on your hand with MediaPipe. Your index fingertip moves the pointer, a pinch of thumb and index clicks, and two raised fingers scroll. It runs in this page, on your camera, and the video never leaves your browser.
| Move your index finger | Move the pointer |
|---|---|
| Pinch thumb and index | Click |
| Raise index and middle, move up or down | Scroll |
| Pinch thumb and middle | Right click, in the desktop app |
91.3%
Class XII, MPC, Velivennu
Mathematics, physics and chemistry.
Final year
B.Tech, Artificial Intelligence and Data Science
Machine learning, data structures, databases, operating systems, networks.
Since 2024
Co-ordinator, IIITDM Kurnool
Covers the major campus events, from the schedule to the final pictures, and runs workshops and competitions.
Recommended
Python Developer Intern, remote
Python and Flask automation and backend tools with SQL, and debugging existing programs with the team.
Best performer
AI and ML Intern
30-plus daily problems in preprocessing, visualisation, scikit-learn and TensorFlow.
4 Superbadges
Salesforce Developer Intern, remote
Eight weeks of Apex, Lightning Web Components, SOQL, data modelling and process automation.
I like problems where the input is messy and physical: a creased medicine strip under bad light, a hand waving at a webcam, an hour of video nobody wants to sit through. I build the whole path, from the model to the app someone opens, and I make it run on hardware people already own.
Point at a tool to see where I used it.
PointNet, trained on ShapeNet, labels every point of a 3D object with the part it belongs to: a chair's back, seat and legs. 2026
Listens, then reads out the weather and the news and sets reminders, with speech recognition and text-to-speech. 2024
Captures the empty room first, then shows it wherever the cloak is, so whoever wears it disappears. OpenCV, on a webcam. 2025
A ResNet-152 reads the picture and an LSTM writes a sentence about it, one word at a time. 2025
Selenium runs a Google search and pulls out each result's title, link and description. 2024
Everything on GitHub 15 repositories
