Samuel Edwards - Software Developer

Experience

Jul 2025 - Present

Remote

CVS Health

Software Developer

Cut pharmacy analytics model execution time 40-60%, holding output within 0.5% of legacy values, by migrating SAS/Teradata models to Python/BigQuery, then Dataiku/BigQuery. The models process hundreds of millions of records.
Closed a data-integrity gap ahead of the Data Engineering team's own tooling by building, shipping and maintaining an AI agent (Claude Code) that catches stale tables and validates BigQuery/Teradata consistency.
Built a custom MCP server that connects Claude Code to the Dataiku platform, so the rest of the engineering team can build against production Dataiku projects with AI coding tools.
Built a self-serve support agent front to back, with a RAG knowledge base, the Dataiku MCP server so it can run models start to finish, and a Streamlit chat UI. It cuts the platform and model questions that reach the Models team.
Built an internal Python library with a BigQuery query API. Non-technical users run cloud-native queries without raw SQL or local compute, and memory overhead dropped 90%.
Defined infrastructure standards and reliability guidelines for the team's Dataiku platform. Other teams on the platform adopted them.

Aug 2024 - Jun 2025

Remote

U.S. Census Bureau

Computer Scientist

Helped modernize file-based data infrastructure onto an AWS Redshift and Python stack, supporting reliable delivery of 200M+ records across 150+ variables on data.census.gov.
Built and published sas-to-polars, an open-source Python library on PyPI, to replace a far slower way of converting legacy .sas7bdat files into Polars DataFrames. Processing time dropped 70%+, and other groups at Census adopted it.
Protected sensitive data in custom tables delivered to clients by injecting statistical noise and running multi-step automated verification, meeting Census data-quality and privacy requirements.

Jun 2024 - Aug 2024

Turlock, CA

Hilmar Cheese & Ingredients

Information Technology Intern

Built hourly ETL pipelines that consolidate employee permissions data from SAP ECC, Azure AD and SAP SuccessFactors into an interactive Power BI audit dashboard, with pipeline monitoring that shows stakeholders how fresh the data is.
Improved identity and access management discrepancy resolution time by 90%+ across 100+ discrepancies with Power Automate and Power Query workflows.

Projects

HAI OS

Node.js, PostgreSQL, Gemini API, Vapi, Docker, GCP Cloud Run, GitHub Actions

A multi-agent support platform. A router agent classifies each message and hands it to a specialist agent (triage, technical or billing), logging the reason for every handoff. Each agent has its own RAG pipeline with Gemini embeddings. System prompts reload without a restart, and staff can watch live sessions and take over from the agent. Voice calls run through Vapi over an OpenAI-compatible streaming endpoint.

sas-to-polars

Python, multiprocessing, Polars, pyreadstat, PyPI

An open-source library that converts legacy SAS files into Polars DataFrames across multiple cores. Written at the Census Bureau, where converting large .sas7bdat files was the bottleneck. Available on PyPI.

Game location and NBA free throws

Python, causal inference

A causal study with a three-person team on NBA game data from 2004 to 2022, testing whether playing away changes a team's free-throw percentage. We found no significant league-wide effect, which corrected our own starting hypothesis.

Skills

Languages: Python, SQL (PostgreSQL, BigQuery), JavaScript, Java, R, SAS, Bash
Data: ETL/ELT, Dataiku (Scenarios and API), BigQuery, Redshift, Teradata, Polars, PySpark, Pandas, MongoDB, REST APIs
Cloud: GCP (BigQuery, Vertex AI Workbench, Cloud Run, Cloud Functions), AWS (Redshift, EC2), Azure AD, Docker, GitHub Actions
AI: Claude Code, MCP server development, multi-agent orchestration, RAG, embedding search, voice AI (Vapi), Ollama, scikit-learn
Tools: Git, Linux, Postman, Power BI, Power Automate, Streamlit, Snyk

Education

May 2024

University of California, Berkeley

B.A. Data Science

Coursework included data engineering, data structures and algorithms, data mining, probability and statistics, causal inference, and machine learning.