aryan raj

I’m an AI/ML and computer vision engineer currently building real-time solutions that work reliably on edge hardware. I focus on translating strong models into dependable pipelines with clean logic, measurable accuracy, and practical performance constraints.

In my internship at Tata Motors (TMCV — Cadre & Training), I engineered a real-time human detection and counting system using YOLOv7-tiny, Raspberry Pi, Python, OpenCV, and PyTorch. I implemented person detection, centroid tracking, ROI-based virtual line crossing, and entry/exit counting logic, reaching 85–90% detection accuracy in good lighting and over 90% counting accuracy in controlled scenarios.

I also optimize for deployment realities—maintaining 5–8 FPS and 3+ hours of stable continuous runtime on Raspberry Pi without GPU dependency. I bring that same mindset to projects like a gesture-control system using OpenCV and MediaPipe for low-latency desktop interaction.

Beyond vision, I build AI-enabled product components: MemoryOS combines embeddings, semantic search, and a local-first memory architecture, while my OCR project uses ChandraOCR 2 to support multi-format inputs and improve structured extraction with layout-aware prompting. I’m also a leader—having grown a student photographic society from member to president and managed a 45+ member team—so I value teamwork, ownership, and results.