https://debounx.com/notes/ Practical notes on AI support triage, catalog enrichment, LLM evals, and internal operations. en-ca https://debounx.com/notes/ai-support-triage-evals.html https://debounx.com/notes/ai-support-triage-evals.html Wed, 21 Oct 2026 09:00:00 GMT Four metrics, an eval set from real tickets, a failure budget, and the handoff that keeps a human on edge cases. https://debounx.com/notes/catalog-enrichment-pipeline.html https://debounx.com/notes/catalog-enrichment-pipeline.html Wed, 21 Oct 2026 09:00:00 GMT Source-of-truth rules, a 200–500 SKU sample batch, per-field accuracy, and a review queue before publication. https://debounx.com/notes/internal-ai-guardrails.html https://debounx.com/notes/internal-ai-guardrails.html Wed, 21 Oct 2026 09:00:00 GMT Least privilege, approval flows, retrieval freshness, audit logs, and adoption metrics for internal assistants.