Tool calling: when you need Jev and when LLM caching is enough
Prompt caching cuts the cost of resending tools and context. It does not make a tool-call decision typed. When a Jev gate earns its place and when it doesn't.
Read the articleBetter Technologies,
Better Results
Notes on the technology decisions behind the systems we build.
DPDP proof packs: artefacts auditors and boards ask for
After the DPDP roadmap, readiness is artefacts you can show. A proof-pack inventory for logs, purpose registry, DSR SLAs, retention, and board-ready evidence.
Moving off a legacy platform without breaking what works
Big-bang rewrites break the paths that still earn money. A safer legacy system migration keeps the old platform earning while you grow a new edge beside it: coexistence, parity contracts, traffic shaping, sync with reconciliation, and rollback that is a config change.
Discovery to hypercare: a delivery shape for AI programs
What good AI project delivery looks like: phases, exit criteria, and ownership from discovery workshop to hypercare and ops handoff. Built to end pilot-forever.
Knowledge platforms that agents can cite, not scrape
Enterprise AI knowledge platforms need retrieval, provenance, and access control so agents cite governed sources instead of scraping the intranet into unofficial policy.