Senior Data Scientist
Core
Lead analytical methods and experimental design to improve the U.S. Census Bureau's LLM Autocoder for coding survey write-in responses.
Role type
Senior IC data scientist (NLP & LLM evaluation)
Builds
Evaluation frameworks and coding methods for the American Community Survey (ACS) Autocoder
Domain
Federal statistics / NLP / Text classification
Deliverable
production ML models
Required skills
Python, SQL, NLP, text classification, LLM application, experimental design, model evaluation, statistical analysis, error analysis, sampling, class imbalance handling
Preferred skills
Advanced degree, survey data experience, industry/occupation coding taxonomies, AWS analytical workflows, federal statistical environments
Technologies
Python, SQL, LLMs, fuzzy matching, dictionaries, rules
Responsibilities
Establish performance baselines and evaluation methods for the ACS Autocoder; Analyze accuracy, coding rates, coverage, and error patterns; Design reproducible experiments with appropriate controls; Compare LLMs and hybrid approaches based on quality and cost; Assess data readiness and coding taxonomies for additional surveys; Guide engineers on prompt and model configuration changes; Interpret results and document technical guidance
Seniority
Senior, hands-on IC