Your question is Dataset Curation for Instruction Tuning. Take a moment with it on the right.
Talk me through your thinking if you like. When you're confident, submit your answer and I'll grade it like a real screen (7/10 or better passes).
How do you design an effective dataset curation and filtering pipeline for instruction tuning a model on specialized customer support data at Decagon?
Describe a practical pipeline for ingesting support conversations, tool calls, agent actions, and outcome signals; removing sensitive or unusable content; validating instruction-response examples; and producing versioned training datasets. Explain how you would support incremental processing, reproducible backfills, schema evolution, data lineage, human review, and monitoring for quality, leakage, duplication, and distribution drift.