AI Systems (2023–2026)

• Created “Zippy”, Zipline’s production RAG assistant and flagship AI feature—found the approach in research, pitched it, built the prototype in two weeks, named it, and shipped it. Live with 50+ retail brands across 1,000,000+ conversations and 2,000,000+ pages indexed; still the product’s primary AI differentiator three years on.
• Built an ingestion pipeline that breaks documents into self-contained “atomic facts”, so each retrieved chunk keeps the meaning of the document around it. Anthropic later published a closely related idea as Contextual Retrieval.
• Re-architected the information retrieval engine, twice. Led the migration from pgvector to Elasticsearch, then designed and shipped vision-model document understanding so Zippy could answer questions about diagrams, planograms and tables.
• Shipped agentic tool use to production, including MCP (Model Context Protocol) integration that lets Zippy query live systems for real-time inventory, scheduling and policy answers rather than only indexed documents. Built the MCP server model, client and admin interface, and multi-step tool calling via the OpenAI Responses API.
• Made retrieval quality measurable. Co-built Zippy’s first LLM-as-judge evaluation system. Built the test sets it ran against, including one generated from three years of real customer bug reports. Migrations were decided on measured pass rates instead of opinion.
• Engineered fast, access-controlled retrieval at enterprise scale: vector search combined with permission-predicate filtering, working around HNSW’s known degradation under predicate constraints so results stay both fast and correctly scoped per user, team and role.
• Built the full conversational front end with Hotwire (Turbo + Stimulus): token-streaming responses, multi-turn conversation state, source citations, query refinement, and inline feedback capture.
• Shipped a multilingual retrieval fix over a weekend to unblock a customer’s national television segment, adding automatic query translation so non-English questions retrieved answers correctly.

Architecture & Technical Leadership (2025–2026)

• Settled architecture arguments with evidence. Ran a study on real customer documents to show which extraction failures actually occurred, and won committee approval for vision-based extraction on that evidence.
• Challenged an infrastructure cost projection by pricing it against the existing pipeline’s actual invoiced spend, then ran the vendor review that surfaced a substantially cheaper path.
• Served as the company’s AI technical authority—standing member of the AI Steering Committee, author of the RFCs setting AI architecture direction, and the escalation point for AI questions from engineering, product, sales and customers.
• Fronted the platform’s AI story externally: enterprise security and responsible-AI reviews, sales RFPs, and vendor architecture negotiations.
• Acted as team lead for the AI/search team: cycle planning, technical direction, and distributing architecture work across the team rather than holding it.

Product & Platform Engineering (2019–2023)

• Built Assessment tools end to end—historic results, auditee search, scoring and sorting, and the complete auditor workflow. 116M+ assessment questions answered.
• Built Report Builder (custom columns, filters, role-based access), Learning Insights learner reports, and discussion threading with notifications and mentions across the messaging system.
• Built platform capabilities: REST Hooks for Zapier integration, a full video transcoding pipeline on AWS MediaConvert with HLS streaming, and MegaSeed—an internal tool for generating large-scale realistic data to test performance as the platform scaled.
• Migrated legacy reporting to BigQuery as part of a small team, matching edge cases exactly so historical numbers stayed correct.
• Kept the application fast as Zipline grew 10x in active users, through Rails and SQL optimization, caching, and automated performance testing in CI.

• Led effort to consolidate logs, metrics and traces across services in large microservice architecture, increasing visibility and reducing MTTR.
• Worked on migrating several services from Digital Ocean to AWS.
• Assisted with book club, leading sessions on designing microservice architectures and object-oriented programming with Ruby.

• Designed and developed a continuous integration/deployment system for the mobile and desktop websites (m.accuweather.com and www.accuweather.com)
• Created a complete monitoring system for the public website, tracking 100s of statistics for 10s of millions of requests daily.
• Created automated systems to deploy most public-facing AccuWeather presences, and ensure operational consistency within clusters.
• Created several new open-source Ansible role libraries, and contributed code to Terraform.
• Created an open-source query editor for the Microsoft Application Insights Grafana data source.

• Designed and developed a dashboard with Javascript and D3 to quickly convey pertinent information concerning precision agriculture.
• Fixed a multitude of bugs within a 10+ year old codebase, building a full–stack testing system to automatically watch several ancient error–prone areas.
• Overhauled user authentication, bringing it up to speed with modern encryption standards.
• Designed and developed several front-end tools, including a parallel-computed variable rate irrigration pivot zone tool (VRI), and a fertilizer blend tool calculator using linear programming.

• Designed and implemented several major service-oriented applications for the precision agriculture industry.
• Devised programming strategies for handling wide variety of raw file data formats.
• Optimized code for handling big data problems (100 million rows per month).
• Rewrote legacy code to distribute work and reduce load (100s of workers, 100,000s of jobs daily).