ROBOTICS SOFTWARE ENGINEER

Where research meets the production line.

At Anthropilot, I turned a state-of-the-art transformer detector, RT-DETR-L, from a pretrained checkpoint into a dependable pair of eyes inside a real-time Python vision runtime. That meant architecting the full inference pipeline end to end: how frames arrive, how masks get cleaned, and how each detection earns the system's trust before it is allowed to act.

The heart of the work is judgment under uncertainty. I designed confidence-gated safe/unsafe classification and mask post-processing logic so the system knows when to proceed on its own and when to raise its hand, and I instrumented run-level telemetry throughout, so deployment readiness is measured, not assumed. It is quiet, careful engineering: the kind that lets a robot make the easy calls itself and saves human attention for the moments that genuinely need it.

From pixels to policies to propellers.

Maryland is where I learned to think like a robot, and then to think better than one. It started with perception: teaching machines to see through cameras, depth, and optical flow. Then planning gave that vision purpose, turning maps into motion with A*, RRT variants, and trajectory optimization. Control theory grounded everything in physics, where LQR and LQG taught me that elegance means stability under noise, not just clean math.

From there the story got more human. Robot Learning showed me how machines improve from demonstration and reward. Natural Language Processing let them take instructions in plain English. Decision making in robotics taught coordination when many robots share one world, and Human-Robot Interaction asked the hardest question of all: how autonomy should behave around people. The final chapter took flight with hands-on autonomous aerial robotics, where every lesson above had to survive on real UAV hardware.

Mechanical first. Software second.

Before I wrote software for robots, I learned what robots are made of. Mechanical engineering gave me the physical intuition every roboticist secretly relies on: kinematics, dynamics, materials, and the discipline of CAD, from AutoCAD to SolidWorks and Fusion 360. Designing and manufacturing a pneumatic vice taught me that tolerances are not suggestions, and investigating electromagnetic shielding with carbon nanotubes taught me how research actually works.

That foundation is why I trust my simulations today. When I export a URDF or tune a controller, I am not treating the robot as an abstraction. I know where the mass sits, why the joint flexes, and what the real world will do to a perfect plan. That order made all the difference.