Data Delivery Scientist
Created production-level Jupyter workflows simulating realistic multi-turn conversations for LLM training and evaluation. Structured agent instructions using JSON and built instruction-following datasets with supervised fine-tuning approaches. Trained and evaluated LLM behavior using adversarial prompts and edge-case task scenarios. • Simulated multi-turn dialogue data for agent development. • Applied SFT and instruction-following modeling concepts to dataset preparation. • Ran LLM training/evaluation cycles using adversarial and edge-case prompts. • Validated model responses for relevance, context accuracy, and intent alignment.