I design, build, and ship production agent systems for the enterprise.Senior Forward Deployed Engineer at Salesforce.
Demos are easy. Production is the job. The gap between a promising agent and a dependable one is where I work.
A production customer service agent, handling more than 120,000 automated actions a week with case deflection between 40 and 49 percent. I led it through the launch failure that nearly ended it, and I'm now leading the build of the multi-agent architecture it moves to next.
Major US airline02Two live agents, one facing diners and one facing restaurant operators, together carrying between 8,000 and 15,000 conversations a week. I rebuilt the escalation path so a handoff is computed from business hours and real availability, not guessed by the model. The operator-facing agent runs at 80 percent deflection, and the agents recorded 1,156 successful customer reactivations.
Global dining platformA partner deal registration program validated in UAT across 17 languages, spanning the US, Japan, and Asia-Pacific. I architected the foundational program.
Global consumer technology companyA large-scale AI proof of concept for a contact center, with a projected material reduction in average handle time. I designed and delivered it, and projected is deliberate: a proof of concept earns a forecast, not a production claim.
National telecom carrierAn AI insight engine prototype my team built for an internal engineering hackathon, named one of five winners out of 63 submissions.
Internal innovationTwo questions I ask on every build, each one turned into something you can run. The design gate decides what to build. The ship gate decides whether it is ready. The write-up behind both is in the two gates, and the whole method installs as a Claude Code plugin, gates plus a loop that turns production incidents into regression evals.
I keep seeing agents built where a workflow would do. Answer seven questions and get an opinionated call on which parts actually need reasoning, and which should stay deterministic automation you can test. Use the browser version, or run the same decision over a real spec from your editor with the Claude skill.
A diagnostic for telling a demo agent from a production one. It scores an agent across seven dimensions, shows where it is strong and where it is thin, and turns the gaps into a plan. Use the interactive version, or run the same review from your editor with the Claude skill.
Live model output. It can be wrong. For anything that matters, email.
I’m a Senior Forward Deployed Engineer at Salesforce, where I lead production AI deployments for some of the company’s largest enterprise customers. My work sits at the intersection of architecture and delivery: multi-agent systems, RAG pipelines, and the evaluation and observability layers that make them safe to run at scale.
Recent work includes taking a major US airline’s customer service agent through the launch failure that nearly ended it, and now leading the build of the multi-agent architecture it moves to next. I also rebuilt escalation for a global dining platform’s two service agents, so a handoff is computed from real availability instead of guessed by the model, and I advise Fortune 500 teams on AI trust, security, and governance. Earlier in my career I shipped enterprise programs across consumer technology, telecom, and travel, from strategy through delivery.
I hold the Claude Certified Architect certification, 17 Salesforce certifications, and a professional certificate from MIT xPRO. Based in Los Angeles.
Production over demosAn agent that works in a keynote is not an agent that works at 2am.
Cost is a featureRuntimes should get cheaper as they get smarter.
Trust is architectureGovernance belongs in the system design, not the slide deck.
Salesforce
Production agent deployments and technical advisory for strategic enterprise accounts.
Salesforce
Architecture and delivery leadership across consumer electronics, telecom, and travel programs, leading teams of 25+ engineers.
Accenture, Deloitte, and Cognizant
Claude Certified Architect, Anthropic · 17 Salesforce certifications · Professional certificate, MIT xPRO