Ai/Ml Intern
Client: AI23Engineered production-grade LLM infrastructure leveraging advanced prompt engineering and few-shot learning techniques, achieving 85% accuracy in converting unstructured data to structured JSON outputs.Optimized large document processing by implementing dynamic chunking strategies and context window management, reducing token usage by 40% while improving inference throughput by 3x.Developed comprehensive human-in-the-loop evaluation framework for LLaMA 3 fine-tuning, incorporating feedback loops and achieving 92% task-specific accuracy through iterative model improvements.