joel brook

• –Independently built and shipped a self-service React feature within a shared suite of internal engineering tools used by hundreds of engineers, owning it end to end from implementation through testing and deployment
• –Integrated internal APIs to fetch valid configuration options and submit generation requests for config files and test/simulation inputs through a guided browser-based workflow
• –Cut a 30+ minute manual command-line process to a few minutes, letting engineers unfamiliar with the underlying CLI tooling generate artifacts without locating, configuring, and running it

• –Joined during a ground-up rebuild of Bloom’s cafe-loyalty platform and designed and built its Flask and MongoDB backend, working full stack across nearly every part of the product
• –Built core flows end to end, including authentication, customer sign-up, and rewards logic, across the customer-facing rewards app and the merchant application
• –Developed Espresso AI, a merchant analytics tool that turned transaction and customer data into loyalty and sales insights for pilot merchants

• –Co-founded and operate a collectibles business generating $70,000+ in revenue, managing $45,000+ in current inventory, and sustaining 80%+ net profit margins through strategic sourcing and pricing
• –Built a full-stack inventory system and customer-facing catalog tracking 100+ products with pricing, details, and images, replacing spreadsheets and keeping private business data separate from the public site

• –Engineered an automated backup system using AWS, Jira integration, and Ubuntu VMs that synced hundreds of internal documents every few hours, replacing a manual engineer-run process and strengthening disaster recovery
• –Researched U.S. cybersecurity compliance requirements and competitive landscape to shape the company’s North American market-entry strategy

• –Built a Python prior-authorization engine that turns TypeSafe Jev’s typed, probability-scored judgments into safe actions, with a versioned policy engine (not the model) choosing auto-approve, request-info, or human review
• –1,000-case holdout: 25% auto-processed with 0 unsafe approvals and 90% correct, vs. 13% and 67% for rules only
• –Designed an evaluation loop (100-case hand-built gold set, trace replay, 20 CI regression gates) that exposed a paused-medication flaw; the fix raised correct actions from 69% to 92% on a fresh 1,000-case holdout
• –Matched Claude Opus on the gold set (91% vs. 93% correct) at ~136x lower cost per case

Python · Pydantic · Typer · TypeSafe Jev · Claude API · pytest · GitHub Actions · Next.js · TypeScript

• –Built a Python Discord bot that uses the Claude API to translate natural-language queries into precise Scryfall search syntax for Magic: The Gathering card lookups
• –Implemented fuzzy string matching for card-name recognition to resolve misspelled and partial queries; added SQLite-backed query logging for usage analytics

Python · Discord.py · Claude API · Scryfall API · SQLite

• –Coursework: Algorithms & Data Structures, Computer Systems, Object-Oriented Design, Database Design, Cybersecurity, Network Fundamentals

• –1st place at BU Invitational, NEIFC Men’s 3-Weapon Club Team Championship, and USACFC Club Nationals

• –Multiple top-50 finishes at national-level events; prepared for Pro Tour events with professional testing teams