Araas Boloorchi

Automated Tire, Inc. — Woburn, MA

A quick understanding, not a deep dive. The platform is proprietary, so specifics stay inside the company. This is the high-level shape of the work.

• Lead perception engineering: sensor selection, algorithm development, and on-robot deployment
• Evaluate and select depth and machine vision hardware for the platform
• Develop vision models that track a compressed tire's geometry during the tire change, with millimetre-scale sensitivity
• Optimise perception to run in real time on the robot

FlyX Technologies — Sunnyvale, CA

• Built autonomous drones that land on power lines using RL policies trained in NVIDIA Isaac Lab with PhysX contact dynamics, carried from simulation to real flight
• Isaac ROS cuVSLAM on Jetson Orin with RealSense D435i — real-time 3D SLAM and localisation at 25 FPS for GPS-denied outdoor navigation
• Extended cuVSLAM with improved IMU–camera fusion, cutting drift 40% and taking it from prototype to production on UAVs
• MAVLink flight control linking Jetson Orin to CubePilot Orange: bidirectional telemetry, RC override, real-time IMU visualisation
• High-fidelity Isaac Sim environments with Replicator and domain randomisation for large-scale synthetic training data
• ROS 2 servo control with PCA9685 PWM for gimbal manipulation; fine-tuned YOLO for aerial detection on edge devices with a CI/CD deployment pipeline

Catch Up AI — San Francisco, CA

• Co-founded the company and led a 4-engineer ML team from concept to production MVP in 12 weeks, establishing ML pipelines, model lifecycle management and CI/CD; secured initial customer pilots
• Architected a multi-modal pipeline with distributed cloud inference, improving precision from 0.71 to 0.86
• PyTorch GNN modelling social interaction graphs, lifting recommendation accuracy 22%

ILead Lab — Stillwater, OK

• CV algorithm (OpenCV, scikit-learn) for automated statistical chart interpretation: 94% detection accuracy, 91% user-rated interpretability
• Explainable AI system with full inference traceability, feeding research on interpretable perception
• Augmentation and architecture tuning (Keras/TensorFlow) raised detection recall 7%

Sanborn — Colorado Springs, CO

• Fine-tuned a stereo depth estimation model, cutting inference latency 30% via TensorRT/CUDA without losing accuracy
• End-to-end 3D reconstruction pipeline with production-grade stereo calibration for outdoor deployment, improving perception accuracy 15%

General Electric — Oklahoma City, OK

• Shipped an end-to-end industrial vision system for automated defect detection in manufacturing, at 88% F1 with model lifecycle management
• Integrated camera selection, lighting optimisation and environmental adaptation for robust perception in unstructured industrial settings

Advanced Technology Research Center (ATRC) — Stillwater, OK

• Novel sensor fusion methodology for visual-inertial odometry integrating camera and IMU data under US Air Force funding, reducing computational complexity 5% while holding localisation precision
• Real-time SLAM prototype for GPS-denied drone navigation in adversarial outdoor conditions

Mechatronics Research Lab

• Programmed Nao humanoids for the RoboCup Soccer World Cup: autonomous bipedal locomotion, real-time perception and multi-agent coordination in dynamic, unstructured environments
• Computer vision and sensor-driven decision-making for object tracking, obstacle avoidance and team-based manipulation