I turn real-world workflows into AI agents that do the work.

AI Solutions Architect & Engineer · BAM

AI-native revenue cycle management (RCM) platform automating eligibility, billing, and prior authorization for specialty practices.

• Architected and built BAM’s multichannel prior-authorization (PA) agent, routing each case across clearinghouse e-filing (Availity X12 278), a payer-portal browser agent (AWS Bedrock AgentCore), AI voice calls, and fax under a deterministic state machine with an append-only evidence ledger (360+ new tests).
• Led a live, BAA-covered pilot with a Texas ENT practice through go-live: resolved 82% of flagged prior-auth requests (56 of 68) in 13 days across 110+ payer calls to 9 payers.
• Built Switchboard, a hold-aware AI voice agent for payer calls (Python, FastAPI, Telnyx, Deepgram, OpenAI Realtime, Gemini Live): paid models stay idle through IVR and hold (54% of call time), for ~5x lower modeled cost per call than commercial voice-agent platforms.
• Designed the evaluation framework for voice and browser agents: a blind two-agent call lab with LLM-as-judge, A/B tests across 19 speech and LLM engines, and cost, latency, accuracy, and AI-disclosure regression gates.
• Integrated ModMed’s certified HL7 FHIR R4 API (SMART Backend Services, JWT client assertions, JWKS) and hardened Availity eligibility and PA integrations (X12 270/271, 278), merging ~13.6K lines to production (FastAPI, Next.js, PostgreSQL, AWS Lambda).
• Defined BAM’s PHI guardrails for AI: PHI reaches only BAA-covered, zero-retention models (OpenAI, Vertex AI) while other models see redacted data; PHI-free logs and URLs; AES-256-GCM storage; a secrets vault with automated TOTP sign-in.
• Shipped a signed, MDM-deployed capture agent to 30 clinic workstations (56 releases), recording ~7,900 hours of consented workflow footage, and built the pipeline that mines it into PHI-redacted process maps and role specs (Gemini 2.5 Pro on Vertex AI, Claude, verifier agents).
• Translated observed clinic workflows into a ranked automation roadmap of 28 evidence-linked issues with minutes-saved estimates that set BAM’s build priorities.
• Prototyped BAM’s first desktop product in 10 days (Electron, React, Gemini) and shipped internal agents: a Slack footage “librarian,” a PR agent with Claude-critic quality gates, and a meeting-to-PR coding agent (Claude Code).

Head of AI Enablement (Consulting Engagement) · RepeatMD

Membership, loyalty, and commerce platform for medical aesthetics practices.

• Owned the AI enablement roadmap for CX and platform operations: ran workflow discovery with support and onboarding teams, then shipped LLM applications, adoption plans, SOPs, and human-in-the-loop review patterns.
• Built the RepeatMD CX Agent (Next.js, TypeScript, OpenRouter tool calling): RAG over help-center and Notion content (pgvector) plus live Metabase and SQL data, embedded in the admin console as a Chrome side panel.
• Architected Platform Agent, a multi-model Python pipeline turning unstructured client inputs (PDFs, spreadsheets, images, Gong call transcripts) into structured app-build specs for ~15 client practices.
• Orchestrated 13 models (8 generation, 5 vision) via OpenRouter with routing and fallback, prompt caching, adaptive chunking, JSON repair, and 8-criterion QA scoring; added a 24-worker Gemini tax-code classifier (US and Canada).

Director of Operations · GreyEye

• Led the AI operations roadmap for local inference, workflow automation, and customer-facing prototypes; ran discovery and rapid POCs that lifted operational efficiency 40% and profitability 15% in six months.
• Deployed local LLMs (Python, Ollama, GPU) that cut inference latency 25%; engineered multi-model consensus across GPT-4, Claude, Gemini, Grok, and o-series models, improving output accuracy 30%.
• Built CI/CD for AI workflows (40% faster deployments, 99% production reliability) and migrated workloads to AWS and GCP GPU environments, cutting infrastructure costs 20%.

Senior Program Manager · SimulTrans

• Owned the cross-functional delivery roadmap for a global quick-service restaurant (QSR) enterprise across 27 countries: $100K+ first-year revenue at a 53% profit margin.
• Integrated GPT-4 into production translation workflows in 2023, cutting turnaround 30% with human review controls; automated quoting, vendor management, QA, and invoicing in XTRF (25% faster turnaround).
• Built analytics dashboards and macros that reclaimed 35% of management time and lifted CSAT 15%; restored a critical client relationship, adding $500K+ in revenue (2019–2020).

Program Manager, Talent & Quality · Welocalize

• Designed live linguistic quality assurance (LQA) processes that raised client satisfaction scores to 95%; automated translator workload distribution (20% faster delivery, 15% less coordinator workload); introduced AI-assisted machine translation post-editing, cutting costs 10%.

Program Manager, IP Translations · MultiLing

• Managed 9,000+ patent translation projects a year for Fortune 100 clients; built IP translation and validation systems that sped processing 90%, raised profitability 36%, and cut errors 25%.

Senior Localization Project Manager · Moravia

• Managed global app, web, and technical localization programs at 45%+ profit margins; QA protocols reduced rework 30%.