I build governed data and AI platforms — lakehouse architecture, RAG and LLM systems, and responsible-AI workflows. 15+ years architecting the data infrastructure behind teaching, advising, and research at UC Berkeley.

Open to senior data & AI architecture, engineering, data science, and leadership roles.

Lakehouse & Data Architecture · RAG / LLM Systems · MLOps & Observability · Responsible AI & Governance · AWS · Azure · Python · Spark

I build end-to-end reference implementations — each takes one architecture or technique from raw data through to a working, governed system. Together they trace one arc: prototype the product → industrialize the data foundation → build the AI systems on top of it.

Flagship builds

• Campus RAG Assistant — production-style enterprise RAG platform with a pluggable, multi-cloud LLM + RAG + vector-store provider registry supporting mixed deployments: (AWS Bedrock + OpenSearch) / (Azure OpenAI + AI Search) / mock mode. LangGraph orchestration, RAGAS evaluation, LangSmith tracing, and full CI/CD via GitHub Actions. Next.js + FastAPI. Documentation →
• Scribe IQ Lakehouse — production-pattern healthcare data lakehouse on Synthea Coherent (FHIR R4). A Bronze → Silver → Gold medallion built twice —
  1. Polars + delta-rs + DuckDB on a local/laptop orchestrated with Dagster, emitting one versioned, test-gated Gold data contract
  2. Microsoft Fabric-native Spark + Delta + Pipelines — Two independent, engine-native implementations converge on the same contract by schema parity, not shared code. Documentation →
• Scribe IQ — governed clinical-documentation AI: note-grounded RAG with enforced citations, structured note generation, and a first-class audit trail. Built entirely on synthetic data (Synthea + public clinical-note datasets). The lakehouse above is the principled rebuild of its data foundation. Documentation →
• Fabric HLS Readmission Lakehouse (landing this week) — Microsoft Fabric-native medallion lakehouse and ML scoring for synthetic hospital-readmission analytics, with an explicit Databricks-to-Fabric pattern mapping. Code is ready; the public repository goes up shortly.

UC Berkeley Educational Technology Services (@ets-berkeley-edu) Contributor to the campus data and learning-platform ecosystem:

• boac — Berkeley Online Advising (BOA), the award-winning academic advising platform
• nessie — data pipeline and analytics engine
• chabot — Lake Chabot, an RTL GenAI chatbot platform for support use cases (limited pilot)
• data-loch — AWS data lake infrastructure for learning data
• cloud-lrs — cloud-based Learning Record Store
• cloudlrs-ingest-microservice — learning-events processing microservice for the LRS
• bcourses-chatbot-poc — GenAI support chatbot for internal training/tutorial purposes

Apereo Learning Analytics Initiative Analytics Liaison and Community Coordinator on the Apereo Foundation Learning Analytics Initiative; contributor to the open learning-analytics ecosystem:

• LearningAnalyticsProcessor — an open-source, Java-based analytics workflow manager running Pentaho-based data-integration + ML pipelines; the first automation of OAAI research.

• 15+ years of experience in Higher Education
• ~10 years at UC Berkeley, building governed cloud-native data and AI/ML platforms — the foundational RTL Data Lake and the data systems behind the award-winning Berkeley Online Advising
• Lead Architect on enterprise data lakes & lakehouse, and built data-mesh architectures supporting domain ownership and seamless connectivity across campus
• Developed a high-throughput multi-tenant streaming platform processing ~5M events/day on average
• Data Science and ML/NLP work supporting research enablement — built an MLOps pipeline for reproducible research
• Now working on the campus's governed GenAI initiatives, building production-style knowledge assistants grounded in institutional data with Responsible-AI audit and provenance trails
• Earlier, at Marist University, led the Gates Foundation-funded Open Academic Analytics Initiative, working with principal investigators to build open-source academic early-alert risk models
• Scaled that research to multi-institution production deployments
• Ten peer-reviewed publications in Learning Analytics and Educational Data Mining

LinkedIn · Google Scholar