Frontend to full-stack AI,
one era at a time.
I move fast across domains. Each role pulled me into a new part of the stack — voice AI, then healthcare AI — and each time I owned more of the product. Here's how I got here, and why I'm built for whatever's next.

AI Engineer with 2.5 years building production AI products end to end: LLM pipelines, multi-agent systems (LangGraph), RAG and real-time voice agents, plus the React, Node.js and Python systems around them. At Diagna AI I built the document-AI platform that processes 1,000+ medical faxes a day for live US clinics; before that I shipped VoiceGenie, a generative-AI voice platform, from 0 to $10K MRR in 11 months. I care about AI that holds up in production: guardrails in code, evals, and measured latency and cost. I learn by building, prefer shipping over endless tutorials, and enjoy taking ownership beyond traditional frontend boundaries.
Built the document-AI platform behind FAXFlo, taking it from a company pivot to live paying US clinics. Owned it end to end: multi-model LLM extraction on AWS Bedrock, the HIPAA-compliant backend and queues, RPA write-back into a legacy EHR, and both React frontends.
- ▹Frontend — Built the complete React 19 clinical platform solo: document inbox with dual-view PDF.js viewer, eFax module, appointment scheduling and confirmation flows, and multi-clinic analytics dashboard — per-clinic feature flags, RBAC routing, and route-based code splitting to keep the platform lean as it scaled.
- ▹Backend API — Architected the core of a HIPAA-compliant platform on Node.js · Express · PostgreSQL: 40+ REST endpoints (JWT + RBAC) over a 25-model schema, and S3 · SQS · SNS ingestion into 7 BullMQ/Redis queues with idempotency, rate limiting and a dead-letter queue.
- ▹AWS Pipeline — Architected a distributed document pipeline (S3 → SQS → Textract → SNS → Bedrock) processing 1,000+ medical faxes daily at 99%+ uptime — with dead-letter queues (DLQ) for poison-message handling, SNS-triggered retries, and Slack alerting on failures.
- ▹AI Classification — Led multi-model integration on AWS Bedrock (Claude Sonnet, Amazon Nova Pro, Llama) to classify and extract structured data from 1,000+ medical faxes a day across 40+ document categories — ~95% accuracy, with clinic-specific prompts, structured JSON extraction and confidence scoring.
- ▹LLM Reliability — Pre-flight token counting, context- and cost-aware model routing, schema-validated JSON with retries for malformed output, and a fallback chain across model families, so one model's outage or bad output never stops the pipeline.
- ▹Workflow impact — Chained AI extraction, patient matching, human review and EHR write-back into one workflow: manual fax triage fell from about 3 hours to 10 minutes a day, and a recurring 2-day backlog cleared.
- ▹SMS Orchestration — Built an AI-driven SMS system across three providers (Twilio · Telnyx · Vonage) handling opt-in/opt-out compliance, appointment reminders, two-way patient messaging, and automated follow-up flows — with provider failover so no single outage breaks delivery.
- ▹Voice AI — Built a Voice-AI outbound scheduling module (VAPI + ElevenLabs) for automated patient calls: appointment confirmations, reschedule handling, and intake — cutting manual outreach by ~70%.
- ▹RPA & EHR Automation — Engineered a full RPA suite (Robocorp · Playwright · Python · FastAPI · Redis RQ) that writes into a no-API legacy EHR (MDLand's iframe-heavy UI: OTP/2FA login, patient lookup, document upload, calendar sync), plus Medplum FHIR sync — referral turnaround fell from 4.2 days to same day. A React Chrome Extension embedded the orchestration UI directly on the EHR page.
- ▹Document Engine — Wrote a custom Zustand + Immer editing engine with field-level diffing and patch-only backend sync; integrated @react-pdf-viewer for S3-hosted PDF preview — converting remote URLs to blob objects for CORS-safe rendering with proper `URL.revokeObjectURL` cleanup — plus multi-page batch processing and ~80% image compression before S3 upload.
- ▹Analytics — Built a HIPAA-compliant PostHog dashboard tracking EMR sync rates, document pipeline throughput, appointment conversion, and AI classification accuracy — the same dashboard used in the investor demo that closed the first paying US clinic.
Sole frontend engineer on VoiceGenie, a generative-AI voice agent platform, and a collaborator on its voice infrastructure. Built the product that took it to $10K MRR and closed the first enterprise customers.
- ▹Product ownership — Designed and built the entire VoiceGenie platform from 0 as the sole frontend engineer: customer-facing dashboard, marketing site, and internal tooling — all shipped solo.
- ▹Campaign builder — Built the full campaign creation flow: configure AI voice scripts, select and segment contact lists, set call schedules, define call objectives (lead generation, sales, appointment setting), and monitor live campaign status per contact in real time.
- ▹Voice configuration — Built a per-campaign voice settings panel: ElevenLabs voice selection, pitch, speaking rate, AI temperature, engagement style, and tone — letting teams tune the AI persona for each use case (cold outreach vs. warm follow-up vs. enterprise sales).
- ▹Post-call analytics — Built a post-call review interface with full call recordings, auto-generated transcripts, emotion detection per call segment, and AI-extracted entities (name, email, phone, address, intent signals) surfaced as structured data — actionable intel without listening to every call.
- ▹R&D & internal dashboard — Built an internal analytics platform for the product and research team: call success rates, conversion funnels, drop-off analysis by script section, and voice model performance comparisons — used directly for product iteration.
- ▹Onboarding UX — Designed and built the end-to-end user onboarding: account setup, workspace config, first-campaign walkthrough, and guided integration steps — reducing time-to-first-call for new customers.
- ▹CRM integrations — Connected HubSpot, GoHighLevel, and Cal.com: contact sync, call outcome and entity data logged back to CRM records, and auto calendar booking on successful calls.
- ▹Script composer — Live @token variable composer (@name, @appointment, @product) that resolves per-contact at call time; no hardcoded scripts — every call personalised dynamically.
- ▹Voice infrastructure — Collaborated on the Twilio voice-AI infrastructure behind 2K+ concurrent calls, and on rate limiting that blocked 15K+ abusive requests a day.
- ▹Payments — Integrated Stripe end to end: plans, checkout and billing state in the dashboard.
- ▹Performance & growth — Cut page load time 30% via code splitting, lazy loading, and caching; grew VoiceGenie from 0 to $10K MRR in 11 months as the only frontend engineer.
Frontend, full-stack, AI — open to everything.
I've shipped in the voice-AI era and the healthcare-AI era. I learn by building, take ownership beyond traditional boundaries, and I'm looking for the next hard problem to own end-to-end.