AI/ML engineer with 5+ years of experience building production-grade AI systems. My core strength is taking conversational AI and RAG from idea to a reliable, scalable and cost-efficient deployment, covering both the model side and the infrastructure side. Delivered measurable results: an 80% reduction in QA workload, 75% HR ticket deflection and 85% faster provisioning. I've also led cross-functional teams and turned complex research into systems that perform under real-world load. Conversational AI, end to end I build the full pipeline, not just one piece of it: - Speech layer: ASR / speech-to-text, speaker diarization, TTS and voice cloning, built for real-time, low-latency use - Understanding and reasoning: LLMs, prompt engineering, fine-tuning, - - conversational memory and tool-using AI agents (LangChain, LangGraph) - Knowledge layer: retrieval over private data so answers are grounded and accurate - Deployment: APIs, streaming, monitoring and evaluation so the system keeps working after launch Production-grade RAG Building a RAG demo is easy. Making it reliable is the hard part. My work covers: - Agentic RAG, Multimodal RAG and Knowledge Graph RAG (Neo4j, GraphDB) - Vector search with Qdrant, ChromaDB and FAISS, including chunking, embedding and retrieval design - LLM evaluation to measure answer quality, catch hallucinations and compare changes objectively - Serving with FastAPI, Docker, CI/CD and MLOps (MLflow), on AWS (SageMaker, ECS) and Azure ML GPU utilization and optimization AI performance and AI cost are the same problem. I focus on getting more out of every GPU: - Profiling and bottleneck analysis with NVIDIA Nsight - CUDA / cuDNN and driver environment management for stable, reproducible inference - Multi-GPU scheduling to raise utilization and cut idle capacity - Inference optimization for lower latency and higher throughput - Right-sizing models and infrastructure to reduce compute cost without losing quality Also experienced in: computer vision (object detection, segmentation, facial recognition) and time series / anomaly detection.