Kannan Balasubramanian

• A generative-AI analytics engine that turns raw service-management data into executive reporting. A six-hour fortnightly process now takes under thirty minutes. In live production use.
• A delivery analytics application — hotspot visualisation, automated service-level alerting, trend reporting — conceived at an internal hackathon and then actually deployed into a live engagement, replacing a manual weekly process.
• Both were built hands-on by the person accountable for the process they replaced. That distinction is the reason they are listed here.

Enterprise delivery has spent decades building the machinery of accountability: service levels with teeth, incident discipline, change control, audit trails, evidence that survives a review. Agentic AI is being deployed into those same estates with almost none of it — and with a common assumption that the model can be asked to police itself.

Context Stack takes the opposite position. Runtime enforcement lives outside the model. Policy is compiled, decisions are reproducible, and every audited decision is bound to a hash of the policy that produced it. The framework is an attempt to give AI actions the same evidentiary standard that enterprise change management has required for twenty years.

The project is also built the way it argues systems should be built. Work is split across separate seats with enforced separation of duties: one specifies but never commits, one implements and is instructed that where the specification and the source disagree the source wins, and a single human ratifies anything irreversible. The governing protocol is in version control. Amendments record what is given up, not only what is gained. Every specification error is numbered, with what caught it, in a published error ledger — the same standard of evidence the framework asks of anyone else.