Every Cognilium AI engineering writeup grouped by subject area. Each hub starts with a foundational guide and continues into the supporting writeups.
Safety stock, reorder points, coverage groups and the buffer the planning engine plans TO but never recomputes.
Which platform carries which capability, what a licence actually covers, and what each release wave changes.
What happens between an emailed quote request and a sent Business Central quote — one queue, identity, line matching, supplier choice and review — and where Sales Order Agent stops.
The mechanism every Optimizer shares: read the system of record, compute the decision, write it back behind a human approval step.
Slotting, batching and routing — what each method does, the data it needs, and where it stops being worth doing.
Where Microsoft's Copilot stops and purpose-built models start — and why the Dynamics channel's advice always ends at 'use Copilot, carefully'.
What happens between a contract arriving and a lawyer seeing it — extraction, category routing, parallel scoring, and the escalation rules that decide which clauses a person must read.
Realised price versus list, discount drift, and finding the leak in posted sales lines.
What a Commerce assortment records — a channel, a product, a date range — and the allocation depth, size curve and store clustering it never derives.
When many agents beat one and when they do not: the decision gate, orchestration topologies, coordination, cost, and how to evaluate the result.
Why production knowledge graphs decay and how to keep them correct: entity resolution, mislink detection, grounding checks, scoring and audits.
How AI agents remember: working, episodic, semantic and procedural memory, the context-window limit, retrieval, and the knowledge graph behind it.
RAG → GraphRAG migrations, hybrid retrieval, security, and the patterns that ship when the vector DB stops being enough.
AgentCore, ADK, framework comparisons, multi-agent orchestration, and the deploy/observability stack production agents actually need.
Production voice systems on Twilio + ElevenLabs + Whisper — latency, compliance, and the architecture that survives a real call volume.
Eval suites, judge loops, smart routing, retry/circuit-breaker patterns, and the day-2 ops that keep a production LLM pipeline honest at scale.
Pipelines that turn unstructured PDFs into validated structured data — extraction, evidence-mapping, cross-document linking and multi-tenancy.