Multi-agent systems that survive production
Why most agent demos never ship, what bounded retries and human approval gates actually cost, and how an orchestrator plus narrow specialists outperforms one large generalist.
Speaking & Teaching
Every talk below is built from production work — the platforms, the failures and the decisions behind them — rather than from a deck assembled for the occasion.
01 — Topics
Each can run as a conference talk, an internal engineering session, or a working workshop with your team's own architecture on the table.
Why most agent demos never ship, what bounded retries and human approval gates actually cost, and how an orchestrator plus narrow specialists outperforms one large generalist.
How HIPAA-aligned handling and electronic visit verification forced discipline that then transferred to operations and logistics. Retrofitting governance costs more than designing for it.
Enterprise SMTP and SMPP: domain alignment, reputation, IP warmup, carrier failover and the backpressure decisions that separate a working lane from a blocked sender.
Speech interfaces fail or succeed on interruption handling and latency. A practical walk through the audio cancellation path that makes a call feel like a call.
Diagnosing a slowing company, finding the one unmade architectural decision, and getting agreement to pay for it before it becomes an emergency.
Operating companies in Canada, the UAE and India. Data residency differences, timezone reality, and why the location of data should be a property of the data.
02 — Formats
The briefing
What shipped, what failed, and what the approval gate cost. Written from production runs across NetWit, Plex Dubai and Wahix.