Intelligence & learning

Status: complete (published 2026-08-19). Scope finalized at Batch C kickoff with maintainer POV; see brief.md, findings.md, vendors.md, sources.md.

Scope inventory (v0)

• Estate: BI and dashboards, notebooks and data science platforms, feature stores, model registries, MLOps
• Agent-era: conversational analytics over metrics layers, LLMOps/AgentOps, eval infrastructure (golden datasets, LLM-as-judge, human review queues, regression gates), trace-to-dataset curation
• Improvement machinery: fine-tuning and distillation pipelines, A/B and canary rollout for agent changes, experiment tracking, classical ML (classifiers, topic models, forecasting) as the deterministic muscle next to LLMs
• Risk: model risk management alignment for regulated environments

Challenged-default candidates

LLM-first analytics vs metrics-layer grounding; buying eval platforms vs building on open tooling. Each track proposes its final list at kickoff.

Files

brief.md, findings.md, vendors.md, sources.md are created from ../_TEMPLATE/ at kickoff.

---

Source: research/R06-intelligence-and-learning/README.md in the evidence repository behind this site.