I ship inside a bank. My agents ship while I sleep.

I deliver and support regulated credit-risk data systems on Snowflake, dbt, Python, SQL, GitLab CI/CD, Control-M scheduling, and a validation framework that has to satisfy auditors as well as engineers.

• Delivered three major consumer-lending features in the second half of 2025, including a legacy pipeline retirement and a FICO 10T data asset for CECL, both on the regulatory deadline.
• Corrected and backfilled nineteen years of historical data (2006 to 2025) with targeted comparisons instead of full rebuilds, and kept reporting uninterrupted through the transitions.
• Contribute to the bank's Data Science and Engineering Guild, where decisions and education about how generative AI is implemented across the company get made.
• Built a reusable validation query that collapsed a multi-layer transformation into one check the whole team and its validators now use.
• Mentored teammates, onboarded new engineers, taught agentic development practices, and wrote dozens of SKILL.md guides for the repetitive processes in my area.
• Delivered modernization work across a reusable authorization view, a certified-source migration, and a 300,000-row seed file promoted into a governed Snowflake external table.

I design and build agentic systems for operations-heavy businesses, from discovery and architecture through deployment. The codebases are large and I develop them nearly single-handedly, with an AI development pipeline I've spent years honing: specs, coding agents, verification gates, and evidence on every delivery.

• Architected a store-close reconciliation platform for a multi-store retailer where Python owns reconciliation and ledgers, models own document extraction and summaries, and people review the exceptions.
• Designed SecondHome, a source-to-score qualification system for distressed-property leads with provenance, confidence, and a written call rationale on every record.
• Run isolated client agent deployments on per-client VPS infrastructure with preconfigured container stacks, separate from the shared-host environments I run for private users.
• Build with a fixed pipeline: outcome and acceptance test first, split deterministic work from model judgment, smallest useful build, agents implement bounded changes, deterministic validation, human review where consequences are real, live readback after deploy, evidence attached to the task.
• Lead one employee and partner with a co-founder who runs sales; I own the technical delivery end to end.

I built the autonomous creative-testing loop and the marketing analytics behind a product launch, and led the team of six that kept it fed.

• Built an end-to-end automated Meta Ads pipeline that generated creative variants, monitored performance, killed poor performers, and iterated daily with a human review step in Discord.
• The launch waitlist campaign produced 40,000 leads and brought cost per lead from about $2.19 down to about $0.29.
• Led a team of six and owned the e-commerce, referral, tracking, and survey automation stack.

A student venture that turned into a real one. Personal-finance software that read your receipts better than anyone else on the market. Serverless on AWS before that was the obvious choice.

• Raised $46,000 in cash and about $60,000 in in-kind prizes, largely through the Stumberg Venture Competition, and led a team of seven developers with hiring and firing authority.
• Built the receipt pipeline: S3 to Lambda to Textract, behind a CNN gate. Fine-tuned a GPT-3 classifier on 40,000 curated items, covering 100+ categories at over 85% reported accuracy.
• Shipped structured JSON output from language models before function calling existed, and did bias-mitigation work on the fine-tuned advice-generation model.

I took a credit-card underwriting data feature from requirements to production in one summer. Then I talked a bank into letting an intern demo a local RAG chatbot.

• Parsed semi-structured Kafka JSON into usable Snowflake views, wrote the Python validation scripts, and cleared the production control gate inside the internship window despite two requirement changes.
• Built a local, references-first benefits assistant on ChromaDB and Falcon-40B, no internet access. Presented it to hundreds of employees, then helped five colleagues deploy it.