Senior Software Engineer & Engineering Lead · Austin, TX

I build systems where healthcare meets scale.

Engineering lead building event-driven, real-time systems at 90M+ user scale — from low-latency scheduling engines to multi-agent AI assistants in production.

90M+

members on platforms shaped

7×

faster booking (12.5s → 1.8s)

+57%

scheduling throughput under load

9+

years in regulated systems

Personal project

CurieAI

Personal project — live in production at app.cura-i.com

A private assistant workspace for grounded chat, document retrieval, and multi-step workflows — named for Marie Curie. I designed and shipped it end-to-end: a multi-user RAG system with smart routing, live data, a tool-calling agent, and multi-agent workflows. Every conversation and document is scoped to your account.

  • Smart routing — Each message is sent to fast chat, document retrieval, or a multi-agent workflow (researcher → synthesizer → reviewer → writer) based on intent, with an optional live SSE trace.
  • Live data — FX, stocks, weather, and news return through a deterministic adapter path that skips the model entirely when intent is clear.
  • Tool-calling agent — A LangChain agent picks web and live-data tools per query; RAG answers come back grounded with cited sources.
  • Multi-user & private — Per-user auth, conversations and Qdrant retrieval scoped by user ID; only you can list, open, or delete your chats.
  • Multi-runtime — Same app runs on Ollama locally, LM Studio on a Mac mini over LAN, and Groq in production on GCP.
FastAPIPostgreSQLQdrantRedisLangChainReact 19GCPGroq

Personal project · Phase 1 MVP

CurieFHIR

Open source — HL7 v2 & clinical notes → validated FHIR R4

An LLM pipeline that translates legacy and unstructured clinical data into validated FHIR R4 resources — with a self-correcting validation loop against HAPI FHIR. Built to turn Mirth/HL7 experience into a reproducible, AI-assisted interoperability system.

Pipeline

  1. Ingest

    HL7 v2 · notes

  2. Classify

    LLM intent

  3. Retrieve

    FHIR schema RAG

  4. Draft

    generate FHIR

  5. Validate

    $validate

  6. Correct

    retry · escalate

  • Live Mirth path — Optional Mirth Connect channels feed real HL7 v2 traffic into the same pipeline used for fixtures and evaluation.
PythonGroqOllamaQdrantHAPI FHIRMirth ConnectHL7 v2Streamlit

Selected work

Experience

  • 2022 — Now

    Teladoc Health

    Sr. Software Engineer · Engineering Lead

    Event-driven availability engine replacing batch updates — near real-time scheduling for 90M+ members. Cut CCM booking 12.5s → 1.8s (~7×) at +57% throughput, error rate under 1%. Led Altera & Athena EHR integrations (FHIR/HL7 via Mirth), LCNS payments (13% → 8% cancellations), and the monolith-to-microservices split.

  • Nov 2021 — Mar 2022

    Intercontinental Exchange

    Software Developer

    Automated the unglamorous — a Ruby gem for global access revocation across SDLC tools, and a Rails app standardizing artifact deploys. Internal tooling finance teams quietly depend on.

  • 2019 — 2021

    Cerner · Oracle Health

    Full Stack Developer

    Modernized critical-care Infusion Management and Medication Administration from desktop RCP to React — ~50% performance gain via batching, caching, and event-based processing, with billing and audit trails for clinical compliance.

  • 2015 — 2017

    Wipro · FedEx Ground

    Project Engineer

    Java/Spring routing and payment workflows on Oracle for FedEx Ground linehaul, plus high-priority incident response across quarterly releases. Before healthcare became the mission.

Writing

All on Medium

Oct 2026

Learning LoRA, QLoRA, and Full Fine-Tuning With MedGemma 4B: From Mac to H100

Aug 2026

I Built an AI Pipeline to Translate Legacy Healthcare Data to FHIR.

Jun 2026

I Took My Personal AI System to Production. Here’s What Shipping Taught Me.

Jun 2026

What Running Out of AI Credits Taught Me About Local Models