Shivam Bangare completed Machine Learning for Automotive Safety

Built a perception validation dossier end to end for a synthetic Aurora-7 detector: defined detection at IoU ≥ 0.5, scored aggregate and VRU-specific metrics, and showed why VRU recall—not headline accuracy—governs safety. The work also sliced recall by lighting, visibility, and rarity to surface night/occluded failures, then translated those misses into ISO 21448 SOTIF chains and a conditional deployment recommendation grounded in the evidence.

• Leakage-Free Metric Framing
• Per-Condition Safety Analysis
• SOTIF Functional-Insufficiency Reasoning
• Evidence-Based Deployment Verdicting