At Scraperrs Lab, I built Pulsse, an on-premise system that analyzes live CCTV streams to measure footfall, returning visitors, dwell time, and staff engagement. I integrated detection, tracking, re-identification, face recognition, and pose estimation into a GPU pipeline.
I also scaled Pulsse to process multiple live RTSP cameras on a single GPU, using batched inference and an event bus for reliable data delivery. I annotated CCTV data in Roboflow and fine-tuned YOLO models for real-world footage.
At Scraperrs Lab, I built an Agentic AI workflow in n8n and a voice-and-text chatbot using Whisper and LLMs. I also developed FastAPI backends for computer vision inference and containerized model deployments with Docker.
My projects include a local desktop AI assistant using LLaMA.cpp and MCP, a YOLOv8 inference API, and a RAG-based PDF query system. I’ve also worked on NLP and machine learning projects, including a conversational agent for mental health support.
