250+ cameras
Real-time video analytics at 250+ cameras
A production DeepStream pipeline that turns 250+ live RTSP feeds into real-time occupancy and safety events on four on-prem GPU servers.
~/ projects
Projects
Production systems, research reimplementations, hackathon builds, and lab experiments. Client work is described without names or code.
4× A100 80GB
Two-stage vision-language inspection pipeline
A two-stage VLM + LLM pipeline that inspects and reviews submissions for a national tourism platform, served on 4× A100 80GB.
SM120
CUDA kernels for NVIDIA Blackwell
Hand-written CUDA kernels for SM120 covering fp16 GEMM, GQA attention with M-RoPE, LayerNorm, and fused UNet upsampling.
8.19x p95
Mem0 paper reimplementation
A local-first reimplementation of the Mem0 long-term memory loop for LLM agents with an 8.19x p95 latency improvement.
~125M params
LLM from scratch
Building a ~125M parameter Python code LLM in plain PyTorch, one concept per stage, from tokenizer to serving.
RootGraph
Graph-based context for AI agents running fully offline on a small Qwen3 model served with vLLM. Built at Graph Hacks (FalkorDB × WeMakeDevs), October 2026.
AEGIS
An agentic evaluation system that uses agents to test and score LLM applications.
SizeWise
Body measurements from phone photos, using SMPL-X body model fitting.
AeroIQ
A concept for drone-based inspection of railway track and overhead electrical equipment (OHE) for Indian Railways.
Edge LLMs and diffusion
Running llama.cpp language models and diffusion models on Raspberry Pi and NVIDIA Jetson.
KJ-Mini penguin robot
A 3D-printed wheel-legged penguin robot, designed in Fusion 360 and printed on a Bambu A1.
ESP32 water tank automation
An ESP32 controller that runs the home water pump based on tank level, so the tank never runs dry or overflows.