# Ryan Hughes > I build systems that scale and teams that ship. Staff-level engineer across infra, AI, and full-stack. A decade of shipping at Google, Snap, and Motional scale — then founding, growing, and selling an engineering consultancy. Lately: AI agents, financial infrastructure, and the hard problems that live at the edge of what's possible. This is the LLM-friendly version of ryanhugh.com — Ryan's personal site. The site itself is an interactive 3D ocean scene; this file is the plain-text summary of who he is, what he's built, and how to reach him. ## At a glance - 250TB+ data processed - 120M files in 72 hours - 2B+ requests/day surface - 99% latency reduction - 15+ engineers hired & led - 16+ clients incl. OpenAI ## Career ### Google — Software Engineer — Display Ads (2019 – 2021) Shipping on a 2B+ request/day surface. - Led urgent ad-iframe XSS fixes on a surface serving 2B+ requests a day — coordinating 10+ partner teams, lawyers, and external companies, ramping experiments 0.1% → 100%. - 15+ projects across security, compliance, and performance. - 90% code coverage, 15+ A/B experiments and holdbacks with custom GoogleSQL metrics. - Stack: TypeScript · Google Closure · C++ · Protobuf · GoogleSQL ### Snap — Senior Software Engineer — Internal Tools (2022) An approval system for 200+ engineering teams. - Built and owned an internal approval system used by 200+ Snap engineering teams. - Eliminated a legacy codebase's tech debt: no TypeScript, no tests, 1k+ lint errors, 40+ stale dependencies — gone. - Mentored two junior engineers from onboarding to independent feature ownership. - Stack: TypeScript · React · Node.js · GCP ### Motional — Senior Software Engineer — Autonomous Vehicles (2023 – 2024) Observability for 1TB+ of AV sensor data. - Owned the observability frontend for terabytes of autonomous-vehicle sensor data. - Cut critical visualization load times by 90%+ — migrated DynamoDB to Redshift, 1000x fewer bytes shipped to the browser. - Identified and drove a 99%+ latency reduction in self-driving sim visualizations via a Postgres → Redshift ETL pipeline. - Stack: TypeScript · React · AWS · Redshift · Postgres ### Fan Pier Labs — Founder & CEO — acquired by Valere (2024 – 2026) Founded it, grew it, sold it. - Architected and led 18+ AI, infra, and full-stack projects for 16+ clients including OpenAI and Rupa Health (https://www.rupahealth.com). - Processed 250TB+ and 245M+ files across 9 distributed ETL and infra projects. - 3 production RAG/AI systems, 11 containerized production web apps. - Hired and led 15+ engineers in 1–3 person pods; built the whole engagement machine — SOWs, sprints, retros. - Stack: OpenAI · TypeScript · Python · AWS · K8s · Postgres · Clickhouse ## Selected projects ### 120M files in 72 hours (50x parallel · ~72h · 120M files) An IP litigation firm shipped us a hard drive: 120 million recursively-compressed discovery documents they needed searchable. Instead of a weeks-long pipeline running on Windows servers, I pushed everything to S3 and built a Fargate cluster running depth-first decompression with SQS as the stack. Fifty parallel tasks chewed through all 120M files in about three days, indexed via S3 inventory for retrieval. Stack: AWS Fargate · SQS · S3 · EFS · Python ### The 1000x dashboard (90%+ faster · 1000x fewer bytes) The BI dashboard downloaded entire datasets to the browser — and crashed on the ones that mattered. I led the frontend migration and drove the API-contract and architecture decisions: DynamoDB out, Redshift's columnar engine in, Fargate services in between. Six engineers, six months, 90%+ load-time improvement on the tooling engineering teams, operations teams, and management used daily. Stack: Redshift · DynamoDB · Fargate · React ### Defusing XSS at 2B req/day (2B+ req/day · 10+ teams · 0 incidents) A long-known XSS vector in Google's ad iframes — on a surface serving billions of requests a day, where any mistake is a headline. Led the fix end-to-end: coordinated 10+ partner teams, sat with lawyers and external teams, and ramped experiments from 0.1% to 100% of traffic. Shipped the patch with zero incidents and no measurable hit to revenue or latency at full scale. Stack: TypeScript · Closure · A/B infrastructure ### Search NEU (67k users · 1.8M searches · alive 10 years) As a student: search every class and professor at Northeastern, faster than the registrar could. Scraped 1M+ pages a day across 10+ sites, indexed into Elasticsearch with sub-20ms responses, ran at 99%+ uptime on AWS. Handed it off to four student founders — it's still running a decade later. Stack: Elasticsearch · AWS · React · Node - searchneu.com: https://searchneu.com - Khoury News: https://khoury.northeastern.edu/ccis-fourth-year-student-creates-web-app-for-course-searches/ - Sandbox handoff: https://khoury.northeastern.edu/khoury-colleges-sandbox-development-team-takes-on-search-neu/ ## Writing In-depth engineering case studies from Fan Pier Labs (the consultancy Ryan founded and sold). Projects he led or oversaw: - [Building a Smarter BI Tool with AI Agents](https://fanpierlabs.com/blog/building-a-smarter-bi-tool) — When a client approaches us with the goal of building "the next generation of analytics," the conversation usually starts with a common challenge — and ends with AI agents. - [Creating a High-Stakes Intake Site in 4 Weeks](https://fanpierlabs.com/blog/creating-a-high-stakes-intake-site) — How we built a mission-critical intake platform - verified data, automated contracts, and an embeddable widget - in four weeks. - [From 300GB of Raw Data to Predictive Insights: Building a Scalable AI Dashboard](https://fanpierlabs.com/blog/from-300gb-to-predictive-insights) — An inside look at how we engineered a scalable predictive analytics platform using AI and robust data pipelines. - [How We Downloaded, Parsed, and Analyzed 14,000+ PDFs with AI](https://fanpierlabs.com/blog/how-we-parsed-14000-pdfs) — How we combined lightweight infrastructure, parallel downloads, and targeted parsing to efficiently extract complex tables from thousands of PDFs. - [Introducing Eames, the Meeting App](https://fanpierlabs.com/blog/introducing-eames) — Never be late to a Zoom meeting again. Eames is a meeting app that helps you stay on top of your schedule. - [Is Your Data Being Scraped? A Real Case Study](https://fanpierlabs.com/blog/is-your-data-being-scraped) — A real-world investigation into web scraping — how we discovered it, what we found, and what you can do about it. - [Lessons Learned from a Critical Data Breach: A Case Study](https://fanpierlabs.com/blog/lessons-learned-data-breach) — A detailed case study of a critical data breach — what went wrong, how it was handled, and the lessons we took away. - [Scaling Real-Time Gaming to Thousands of Concurrent Players](https://fanpierlabs.com/blog/scaling-real-time-gaming) — How we helped our client scale a real-time gaming platform to support thousands of concurrent players. - [SearchNEU: The Modern Alternative to Northeastern's Course Catalog](https://fanpierlabs.com/blog/searchneu) — How our founder built SearchNEU, a search engine that grew to 50k+ users and became a staple at Northeastern University. - [When "Just Unzip This" Turned Into a Full-Scale Engineering Project](https://fanpierlabs.com/blog/when-just-unzip-this) — How a half-terabyte of deeply nested archives turned into a 27TB, 120-million-file engineering challenge. - [Wrangling a 100TB Data Set for Machine Learning](https://fanpierlabs.com/blog/wrangling-a-100tb-data-set) — How we built cloud infrastructure to process 100 terabytes of data (100 million images) in just 12 hours for machine learning training. ## Contact - Site: https://ryanhugh.com - Email: ryanhughes624@gmail.com - GitHub: https://github.com/ryanhugh - LinkedIn: https://linkedin.com/in/ryanhugh ## Personal Off the clock: sailing, skiing, climbing, and an unreasonable amount of pickleball. --- There is also an AI twin of Ryan on the site that can answer questions about his work in his voice — ask it anything at https://ryanhugh.com.