I'm Director of Agentic Systems at IdeaRoom, where I lead a B2B SaaS company's shift to an agentic-first operating model from the Executive Leadership Team. 6+ years of full-stack engineering — Verisk, ReliaQuest, IdeaRoom — and a production agent that 50+ people now use as a daily driver. I also consult: helping small business engineering and product teams get fluent with LLMs and agents.
My path to software began in biology labs, automating data analysis with Python. That foundation in complex systems thinking now shapes how I approach AI integration: understanding when emergent behavior adds value and when it creates unpredictable risk. Not every problem needs AI; recognizing the difference is half the battle.
At Verisk, I led the QA org's AI tool evaluation and co-architected a department-wide automation app that saved 30+ minutes a day per user across the team. At ReliaQuest, I built AI agent systems into production workflows and ran the weekly upskilling sessions that brought senior and mid-level engineers along. The pattern repeating across both: the hard part isn't the model, it's getting a team to actually adopt and trust the thing.
That work led to a short consulting engagement with IdeaRoom — designing a hybrid deterministic-agentic system to automate their support pipeline — which converted into a director role on the Executive Leadership Team. I now own end-to-end accountability for IdeaRoom's agentic infrastructure, including the platform operations agent that's delivering 35x throughput and ~97% cost savings on Tier 1/2 customer support work.
Alongside that, I keep an independent consulting practice for small business engineering and product teams. My focus there: helping teams that don't have a dedicated AI org get fluent with LLMs and agents — discovery, scoping, architecture, and writing code alongside them. The lessons compound in both directions.
Current Focus
Agentic Systems in Production
Architecting and operating agentic systems that survive contact with real users — LangGraph orchestration, structured output contracts (Pydantic/Zod), and Langfuse-based observability. The metrics I care about are throughput, cost per outcome, and what people actually use six months in.
Executive Agentic Strategy
Defining company-wide agentic strategy from the Executive Leadership Team at IdeaRoom — Platform AI for customers, Operational AI for internal teams, and the unit economics that decide which bets are worth making.
Team Onboarding to LLMs & Agents
Helping engineering and product teams get fluent with agentic systems — through internal training at the companies I work in, and through consulting engagements with small businesses that don't have a dedicated AI org yet.
AI Implementation Discovery
Figuring out which problems are actually worth throwing AI at, and being honest about the ones that aren't. Most of the value is in scoping and sequencing, not in the model choice.
How I Work
- I ask 'what problem are we solving?' before I ask 'what model should we use?' Most failed AI projects skip this step.
- I test against real outcomes before scaling anything. A demo that impresses leadership is not the same as a tool people use.
- I build the error handling and oversight layer before the flashy parts. Production AI that fails silently is worse than no AI.
- I write code alongside the teams I advise. Consulting advice from people who stopped building tends to age badly.
- I treat onboarding as part of the deliverable. The agent isn't done when it works; it's done when the team trusts it and knows when not to.
Want to talk about a project?