AI/ML Engineer at Cloudera

Fine-tuned Llama-3 and Mistral using SFT, DPO, LoRA, Unsloth, and Hugging Face TRL, improving ROUGE-L and BERTScore F1 by 18% on enterprise datasets. Built and deployed a production Agentic RAG platform using Weaviate, BGE Embeddings, hybrid retrieval (BM25 + dense search), and cross-encoder re-ranking, improving retrieval accuracy by 16% across 3 enterprise tenants. Deployed vLLM inference services on AKS with INT8/FP8 quantization, KEDA Event-Driven Autoscaling, and GPU optimization, maintaining <290ms P95 latency while reducing tail latency by 40% under peak traffic conditions. Scaled distributed model training workloads on Azure Machine Learning, implementing PyTorch FSDP2, ONNX Runtime Optimization, and MLflow Experiment Tracking, reducing model iteration cycles by approximately 30% across 8+ training experiments. Developed Multi-Agent LLM Workflows with structured outputs and internal Judge-Evaluator Frameworks for prompt optimization, response validation, and AI safety guardrail