4 of 4 projects
Real-time facial-emotion recognition, quantized for the edge
Reads emotion from a face in real time on a Raspberry Pi 4, using INT8 quantization to run entirely on the edge so nothing ever leaves the device.
Automatic beat-matched mixing, driven by a learned model
Blends tracks into one continuous mix, using a dual-stage model to pick the technique and generate the automation curve for each transition.
Real-time audio-reactive graphics from live sound
Turns live audio into reactive graphics: a 44.1 kHz ingestion pipeline drives 12+ styles on a 2-D canvas, held at a steady 60 FPS.
10,000+ people to live ICU beds and oxygen
A platform I founded during India's second COVID wave that connected 10,000+ people to live ICU-bed and oxygen availability.
A scientific-computing language with a real compiler: parser to AST, LLVM IR via llvmlite, JIT to native code, plus symbolic and automatic differentiation.
USF Math & Statistics research: a Python pipeline that computes exact combinatorial counts up to a ~602,000-digit result in under 3 seconds, replacing brute-force enumeration (11 s) with an optimized Transfer Matrix Method (0.24 µs), with WebGL2 shaders and matplotlib validating the structures.
A PPO agent trained in a custom Gymnasium simulator to pick the best seat in a room of introverts, with domain randomization across 128 layouts and an evaluation suite measuring 85% confidence intervals on unseen layouts.
A growing set of Python scripts for the calculations I reach for most.