| colorlinks |
|---|
true |
- Data: Building Ground Truth, Prompt Engineering, Evaluation
- Engineering: Data modeling, inference server, REST API frameworks
- Tools: Pydantic, Pandas, FastAPI, Flask, Django, HTMX, Bootstrap, SageMaker, Docker, Jenkins, Kafka, Grafana, Splunk, Embedding DBs (Weaviate, ElasticSearch), LangSmith, Sklearn, InstructorAI, PyTorch, Langchain
PhD in Geography, at SDSU and UCSB, 2015. Data science for social science problems.
Built https://nobsmed.com/ for users to search for biohacking experiences found in clinical studies and Reddit comments.
- ETL of 100,000 studies and 100,000 Reddit comments.
- Customer discovery and user interviews.
- Data science model to parse studies and comments.
- Full stack AI development. (Weaviate and Opensearch, Prompt Engineering)
- The producer of Law and Order is making a murder mystery game. I designed and built the story generator. I started from the primitive concepts of organizing the narrative plot structure around a causal DAG.
- NED (Name Entity Disambiguation) of vendor receipts fed into AMD's sales analytic pipeline. Used LangSmith for evaluation.
SimpleLegal is a legal billing analytics company.
- launched a previously stuck AI feature to flag invoice billing discrepancies
- trained 8 transformer models for 8 billing flag labels
- redesigned and managed 5 workers on the annotation pipeline of the labels
- built the inference server (SageMaker, PyTorch).
Sight Machine is a manufacturing analytics company.
- led backend engineering on biggest public facing analytic feature
- coordinated QA process with sales and engineering
- built company's first distributed tracing
- containerized frontend engineering development environment
HiQ Labs was a people analytics company.
- led the migration from a monolith to a micro-service paradigm
- built the scraping system
- supported data science team
Urban Mapping provided geospatial analytics to Tableau.
- CI/CD/QA pipeline
- Observability
- For a legal ops company, I discovered their training data was wrong and rebuilt their ground truth as a team effort to launch a new language AI billing feature.
- For a gaming startup, I designed and built their murder mystery story generation service, starting from a greenfield design of the primitives.
- For a semiconductor analytics startup, I designed a new computer name disambiguation approach for AMD's sale analytics pipeline.
- For the Langchain open-source library, I contributed an experimental CPAL Chain - first causal LLM chain. Tweet by LangChain.
- For a people an startup, new microservice paradigm.
- For academics, new location based human inequality metrics in Assessing Inequality using Geographic Income Distributions
- For several companies, I built observability systems.