AI Software Engineer | Tecvison (LLM systems, fine-tuning, and RAG evaluation)
Built and deployed an LLM-powered Retrieval-Augmented Generation (RAG) system that selects and uses relevant document knowledge to improve agent performance in real time. Fine-tuned an open-weight Mistral 7B model using QLoRA on proprietary datasets to adapt it for a specific downstream task. Developed LLM evaluation pipelines with custom metrics to rate generation quality and reliability across multiple production models. • Used QLoRA fine-tuning on proprietary training data • Implemented RAG retrieval and context surfacing from documents • Evaluated model outputs using faithfulness, groundedness, and latency metrics • Operationalized and maintained multiple production model versions for consistent task performance