Shafahad Khan Mohammed

Developed a Python-based enterprise RAG platform indexing 850K+ documents for 12K daily queries, improving answer accuracy from 43% to 89%. Built a hybrid retrieval pipeline (FAISS, BM25, cross-encoder reranking) to reduce P95 latency from 2.4s to 0.6s and increase retrieval precision by 38%. Optimized LLM routing across GPT-4 and GPT-3.5 to reduce inference costs by 42% while maintaining quality. Created large-scale ingestion pipelines with semantic chunking to preserve table and code context, and evaluated multiple embedding models/LLMs to balance relevance, latency, scalability, and cost. Implemented multi-agent AI systems using LangGraph and Python and built secure microservices with FastAPI, Redis, RBAC, and SSO integration. Deployed containerized services on AWS EKS/Kubernetes with observability and high availability (99.93%).

Developed real-time perception models in PyTorch for analytics over 500K daily vehicle events, improving detection accuracy by 18% and reducing false positives by 22%. Optimized inference with TensorRT/ONNX/CUDA to cut end-to-end latency by 63% and enable 45 FPS while reducing edge inference costs by 35%. Enhanced multi-object tracking and anomaly detection across 100+ cameras, improving tracking accuracy by 21% and event detection recall by 15%. Built CNN/Transformer-based architectures (including DETR/YOLOv8) and trained/fine-tuned models on multi-million annotation datasets using augmentations and synthetic weather. Designed microservices for scalable edge-to-cloud deployment, implemented MLOps with CI/CD and validation, and delivered high availability (99.5%).