Embedded Systems · Computer Vision · Edge AI
Embedded Software Engineer
I make cameras, LiDARs and IMUs make sense — in real time, on hardware you can hold in one hand. From bare-metal C to neural networks, with three years of field-deployed systems behind me.
// about
I spent three years at Tonbo Imaging building mission-critical embedded systems for defence-grade optics — firmware architecture, multi-sensor payloads, and cameras that had to survive the field, not just the demo.
Now I'm pursuing an M.Sc. in Computer Engineering at Paderborn University while working as a computer vision Werkstudent at Heart Wired Minds, building a stereo vision pipeline that tracks 3D pose and motion of competitive mountain bikers in real time — camera sync, IMU fusion, and inference tuned for embedded hardware.
The through-line: I like the layer where software meets physics. Registers, sensors, timing budgets — and the models that turn all of it into understanding.
/* harsha.h — quick reference */ typedef struct { char *background; char *now[3]; } engineer_t; engineer_t harsha = { .background = "3+ yrs embedded systems —" "firmware, multi-sensor payloads," "defence-grade imaging", .now = { "Stereo vision for MTB biomechanics", "LiDAR SLAM for UAV navigation", "Real-time CV on Jetson & RPi", }, };
// experience
// projects
Fully on-device people counting and demographic analytics with dual CSI cameras — YOLOv5n detection, dwell-time tracking, age/gender estimation and a Flask dashboard with occupancy heatmaps. Zero cloud dependency.
Master project: benchmarking FAST-LIO2 and GLIM on solid-state LiDAR data with EVO trajectory evaluation against ground truth — groundwork for autonomous flight in GPS-denied environments.
Real-time 3D pose and motion estimation for competitive mountain biking: synchronized global-shutter stereo cameras, IMU fusion, and depth inference optimized for embedded deployment.
The detection engine behind the analytics platform: YOLOv5 person detection with centroid tracking, persistent IDs and age/gender estimation in a single real-time pipeline with CSI camera acceleration.
Capstone: an end-to-end system that reads numeric digits through an OpenCV recognition pipeline and translates them into PWM signals controlling motor RPM — closed-loop actuation driven purely by sight.
// skills
// publication
// contact
Open to collaborations and opportunities in embedded AI, robotics perception and computer vision — especially anything that has to run fast on small hardware.
Paderborn, Germany · harsha.reddy1611@gmail.com