Nihar Sojitra completed Machine Learning for Automotive Safety

Built a full perception-validation dossier for a synthetic Aurora-7 detector: defined IoU-based detection matching, scored aggregate precision/recall plus VRU false-negative rate, and sliced VRU recall by lighting, visibility, and rarity to expose the worst operating conditions. The final recommendation was evidence-based and conditional, tying the release decision to daylight-only operation, sensor-fusion fallback, and targeted mitigation for night, occlusion, and rare-class pedestrian risk.

• Leakage-Free Detection Definition
• Safety-Critical Metric Selection
• Per-Condition Failure Triage
• SOTIF Functional-Insufficiency Reasoning