Why My AI Service Advisor Forgot Everything
A practical memory strategy for LangGraph workflows, using an automotive service advisor to explain conversation context, execution state, persistent knowledge, and memory governance.
Trust and control
AI Governance
AI governance should not live in a policy document alone. It has to be designed into architecture, evaluation, operations, and delivery.
Evaluation loops and measurable behaviorHuman oversight, escalation, and auditabilityResponsible memory, data access, and model operations
Related articles
Research notes in this topic
From Automotive Diagnostics to Banking Complaints: What I Learned Building a Local LLM Evaluation Platform
A second domain experiment showing why workflow behavior, retrieval shape, and reusable evaluation architecture matter more than simply adding more context.