Vansh Chauhan
I build AI systems that solve real-world problems.
New to Contra
Vansh is ready for their next project!
Alzheimer's MRI Classifier — Clinical-Style AI Diagnostic Tool
Built an end-to-end Alzheimer's detection system that classifies brain MRI scans into four stages — Mild, Moderate, Very Mild, and Non-Demented — with 99.38% test accuracy.
The model runs on EfficientNet-B0, trained on MRI image data and deployed through a clean clinical-style web interface. Upload a scan, and the system preprocesses it, runs inference, and returns a prediction with class probabilities in real time.
Tech used: Python · PyTorch · EfficientNet-B0 · Computer Vision · HTML/CSS/JS
Applications: Medical AI tools · Clinical decision support · Healthcare diagnostics
1
9
PDF Voice Reader — Karaoke-Style Document Reader
Built a web tool that turns any PDF into an immersive read-along experience. Upload a document, and an AI voice reads it aloud while the text highlights in real time — synced word by word, like song lyrics on a karaoke screen.
No more passive reading. The tool extracts text from the PDF, feeds it to an AI voice engine, and tracks the playback position to highlight exactly what's being spoken as it's spoken.
Tech used: Python · PDF extraction · AI Text-to-Speech · JavaScript · HTML/CSS
Use cases: Accessibility tools · Study aids · Document review · E-learning platforms
1
20
Sentinel — Real-Time Violence Detection System
Built a real-time violence detection system that identifies threatening behavior from live video feeds using pose estimation and deep learning — no object detection, no raw pixel processing.
The system uses MediaPipe to extract skeletal keypoints frame-by-frame, then feeds motion sequences into a BiLSTM model to classify behavior as violent or non-violent in real time. Because it works on pose data rather than video pixels, it stays lightweight and privacy-conscious.
Tech used: Python · MediaPipe · BiLSTM · OpenCV · TensorFlow/Keras
Applications: Smart surveillance · Campus security · Public safety systems
1
23
Built this: an AI-powered study group finder that actually understands how you learn.
Most platforms just match by subject. That’s surface-level.
This goes deeper — matching students based on:
• learning style
• schedule overlap
• difficulty level
• campus context
→ Result: groups that feel right, not random.
Key features:
– AI compatibility scoring (so you know why a group fits)
– Live study rooms with chat + notes + whiteboard
– Smart group discovery (DSA, OS, Discrete, etc.)
– Weekly planner based on your exam + weak topics
– Gamified progress (StudyScore, streaks, accountability)
The idea was simple:
Stop forcing people into groups.
Start engineering the right ones.
Designed for students who don’t just want to study more —
but study better.
Would love feedback — especially on the matching logic and UX.
1
58