Selected work - AI systems that earned their keep.

MyMethod runs voice AI agents for client organizations. ESARC has been a senior engineer on the platform, owning the orchestration layer, the routing logic, the eval pipeline, and the transcript-to-analytics path.

The work is multi-tenant by design. Each client org gets its own agent persona, prompts, and routing rules, sharing the underlying infra and the same eval rigor.

Springhouse is an LLM-driven food planning product. The brief was to take the "what do I cook this week" problem and answer it with real recipes, real macros, and real shopping lists that respect a household's constraints.

ESARC owns the backend. That means the agent orchestration, the recipe ingest pipeline, the calorie math, and the API the app talks to.

I joined Meta Superintelligence Labs as a contract AI software engineer through ESARC. My remit sits across tool-calling, MCP integration, and agent evaluation workflows for next-gen conversational AI.

A lot of this work is under NDA. What I can say is that it lives where research meets product, and the bar is "ship something other engineers will rely on next quarter."

ClearProp is what I build when I build for myself. It is a digital pilot logbook for general aviation pilots, designed around the boring parts of the job: getting your hours in correctly, tracking currency, and producing the export your medical examiner or chief flight instructor actually wants.

It's a side project today and a product I plan to commercialize. It also doubles as ESARC's internal proving ground for new patterns before they show up on client work.

I led the design and launch of Sidney Voice AI at Stuf Storage as the Lead AI Software Engineer. Sidney is a voice agent that handles inbound calls, books tours, and triages questions for storage operations that were drowning in repetitive phone work.

The public moment was ISS Vegas, the industry's biggest trade show, where Sidney generated 150+ qualified leads from a single booth demo over a few days.

McGraw Hill is one of the named clients from ESARC's consulting era, alongside NextDay AI and Just a Drink. I was engaged as a senior AI engineer on content and learning-system problems where LLMs were starting to look useful and the publisher needed someone who had shipped them before.

I'm keeping this study short on purpose. The work is publishable in outline. Specifics are something we can walk through on a call.

Scrubs Co-Pilot is a clinical documentation product I co-founded and shipped from zero to 100+ monthly users in three months. The pitch is simple: clinicians spend hours every day writing notes, and almost none of that time is the part they trained for.

The product ingests audio from a consult, transcribes it with Whisper, and runs it through an LLM+RAG pipeline that produces a structured note in the clinician's preferred format. The clinician reviews and signs. That's the loop.

I spent three years and change at Amazon as a Software Development Engineer working on services in the shipping cost path. The systems were Java, the scale was high, and the work was a mix of "make this faster" and "make this cheaper."

The three pieces that show up below are the ones that mattered most: a data-map redesign that cut latency and cost together, a price-prediction pipeline in Python, and a legacy-to-Native-AWS migration that meaningfully shortened cart times for customers.