Software Engineer Intern – Real-Time AI Systems

• Built a real-time SLAM pipeline combining monocular VO and depth estimation, achieving ~20–25 FPS on noisy indoor data
• Implemented RANSAC-based feature matching (2500 matches, 2000 inliers), leveraging IMU to improve motion consistency under noise
• Resolved scale ambiguity using sparse depth and fused VO with GPS in GTSAM, reducing trajectory error from ~95 m to ~1.2 m ATE
• Debugged sensor noise, calibration drift, and VO failures, improving tracking stability and reducing dropouts by ~30%