Abhishek Gaur

Architected an internal agentic AI platform consolidating 15+ tools into a conversational interface, deployed Llama-3.1-70B on-premises with sub-2.5s p95 latency and reduced inference cost 60%, and established LLMOps observability to cut root-cause analysis time from hours to minutes.

Architected a behavioral analytics platform processing millions of events daily and built multi-model forecasting and anomaly detection pipelines; led multi-turn LLM fitness coaching assistant using RAG and agent routing with sub-2s p95 latency.

Built real-time 3D body pose estimation and temporal pose models for the Peloton Guide, achieving 4× model size reduction via INT8 quantization and TensorRT while maintaining >30 FPS and reducing error rate 35%.

Ported face detection and landmark models to TensorFlow Lite with INT8 quantization, reducing model size 75% and latency 40%, and designed safety-critical in-cabin child presence detection targeting >99% recall for automotive deployments.

Developed AI-assisted video annotation with model-in-the-loop active learning reducing annotation time ~50%, built 3D vision systems for industrial inspection, and implemented pruning/quantization pipelines reducing inference cost up to 3×.

Completed a Master of Science in Computer Engineering with coursework and projects focused on machine learning, computer vision, and systems for deployment.

Completed a Bachelor of Science in Computer Science with foundational studies in algorithms, programming, and software engineering.