ML / AI engineer. I build retrieval and forecasting systems, then measure them hard enough to find out where they don't work.
📍 Denver, CO
| 🔍 Citera — agentic RAG, 733 technical PDFs ✅ 0.96 groundedness, 0.94 recall@5 (n=100, bootstrap CIs) ✅ Ablation found the verifier added nothing — cut it    |
📦 Demand Forecasting & Inventory Opt. — M5, 30,490 series ✅ Held-out WRMSSE 0.6475, ~40% > seasonal-naive ✅ Newsvendor policy: 42% cost cut, fill rate 61%→92%    |
| 🧠 Self-Healing Data Platform (in progress) — multi-tenant pipeline orchestration + LLM diagnosis ✅ Real resilience: per-step retries, per-pipeline circuit breakers ✅ LangGraph agent diagnoses failures via pgvector RAG — recommends, never auto-executes  |
🧪 TestForge — Jira tickets → runnable test suites ✅ Runs generated tests for real, doesn't take the LLM's word ✅ Auto-detects & retries malformed generations   |
| 🌐 GDELT Global Events Lakehouse — streaming data engineering ✅ Ingests a feed that updates every 15 min, 61 raw columns ✅ Bronze→Silver→Gold, 7 build phases complete    |
📚 Lossless Learning — serverless ML-education platform ✅ 150+ curated topics, 5 separately-deployed FastAPI microservices ✅ Vertex AI RAG search, Terraform-provisioned GCP stack   |
| 📣 Blitz (collab) — URL in, marketing package out ✅ 6 sequential LangGraph agents, shared context via ChromaDB ✅ Streams results live over SSE |
🏠 LodgeZilla (3-person team) — lodging marketplace ✅ Seeded from real Inside Airbnb data ✅ Full K8s deployment, HPA autoscaling   |
| 📢 Ad Pulse — ad-serving platform ✅ 6 independent microservices, Go ad server ✅ Helm-deployed, per-service CI/CD   |
🚕 Uber Rides Medallion ETL — Databricks pipeline ✅ Spark Structured Streaming ingestion ✅ SCD Type 2 on every dimension via dbt    |
| 📊 ScaleFlow (team project, CU Boulder) — supply-chain risk platform ✅ XGBoost + SHAP risk models ✅ LangChain/GPT-4o Q&A over Pinecone  |
🤖 BigData Platform + AI Observability Agent — GKE data platform with a RAG diagnosis chatbot ✅ 7 Airflow DAGs, OpenLineage/Marquez data lineage tracking ✅ Chatbot diagnoses pod/pipeline failures via ChromaDB RAG + Gemini  |
Python Go LangGraph LightGBM/XGBoost FastAPI Docker Kubernetes Airflow dbt Terraform Kafka AWS GCP Databricks
How I work — commit to the decision rule before running the experiment; measure the noise floor before claiming a metric moved; delete the component the numbers don't justify.
Open to: ML Engineer · AI Engineer · Data Scientist · MLOps