# Martin Brian · martinbrian.com > Personal site of Martin Brian — engineer, builder of AI tooling, author of The Loop (a daily AI/LLM developer-news digest). This file follows the [llms.txt](https://llmstxt.org) convention so LLM agents can navigate the site without scraping. The companion [/llms-full.txt](https://martinbrian.com/llms-full.txt) ships every portfolio section and every newsletter issue in full markdown for one-shot ingestion. Both files are generated at build time from the live site data. ## Long-form ingestion - [https://martinbrian.com/llms-full.txt](https://martinbrian.com/llms-full.txt): every substantive page concatenated as full markdown — portfolio first, then the newsletter archive (latest first). Use for full-context retrieval; use the index below for orientation. ## The Loop · daily newsletter - [https://martinbrian.com/blog](https://martinbrian.com/blog): All issues. Latest featured at the top; archive below. Discovery footer links to RSS, colophon, llms.txt, and llms-full.txt. - [https://martinbrian.com/blog/rss.xml](https://martinbrian.com/blog/rss.xml): Standard RSS 2.0 — daily-cadence consumption for feed-reader users. - [https://martinbrian.com/blog/colophon](https://martinbrian.com/blog/colophon): How The Loop is made — pipeline overview, AI provenance disclosure, persona/voice design, source list, bet-tracking model. - Tag landing pages (`/blog/tags/`) are static-generated at build time from the union of all post tags. None yet — no issues published. ## Portfolio · about Martin Brian - **Role**: Senior AI Engineer. Tagline: "Let’s build something that ships." - **Pitch**: I ship agentic AI that engineering teams actually adopt. Not demos. Not pilots that quietly die. Things that stay shipped. ### Experience - **Now · Marvik → dLocal** — Making AI do the boring parts. Embedded in dLocal's AI team, where I ship agentic systems on a Claude Agent SDK / Claude Code base. The one that rolled out across the engineering org drives 40%+ end-to-end automation of dev tasks on a high-availability fintech platform. People stopped asking "does this help?" and started asking "how did we work without it?" *(Tags: Claude Agent SDK, Python, Metaflow, Kubernetes)* - **Four years · Ulta Beauty · Innovation** — AR, AI, and a Fortune 500. Contributed to GLAMlab — Ulta's AR/AI virtual try-on and its first public generative AI app. Built a video analysis engine that watches affiliate and influencer content for brand-guideline violations at scale. Plus supporting iOS, web, and backend work across innovation projects. *(Tags: Google ADK, Computer Vision, Swift, React)* - **Alongside · BroLi · Co-founder** — Running a bookshop with code. Co-founded and led engineering at an independent bookshop that applied blockchain and AI to a category that usually ignores both. Owned the stack end to end, led a small team, learned that startups are 20% code and 80% everything else. *(Tags: Blockchain, NLP, Team Lead)* - **Earlier · Universidad de Montevideo · Octobot** — Teaching, shipping, learning. Taught innovation and life-design workshops at UM's corporate innovation lab. Before that, shipped Django/React web and Swift/SwiftUI iOS at Octobot. The teaching part stuck — I still think of great engineering as mostly good explanation. *(Tags: Teaching, iOS, Full-stack)* ### Skills - **I think in**: Python, Swift, TypeScript. - **I build agents with**: Claude Agent SDK, Claude Code, LangGraph, Google ADK. - **I make AI useful with**: RAG, Knowledge Graphs, NLP, Computer Vision. - **I ship on**: Kubernetes, Metaflow, Docker, Vercel. ### Projects - [Seniorify.dev](https://teams.seniorify.dev/) — 🏆 1st place at the Montevideo edition of Vercel's "Zero to Agent" Hackathon. A senior-in-the-loop for AI coding agents — intercepts PRs before they're written and interrogates juniors on their architecture choices, so they understand the why and not just the what. *(Tags: AI, DevTools, EdTech, Next.js)* - **Phex** — 🏆 1st place at the 2025 AI Hackathon (Microsoft AI Co-Innovation Lab × INGENIO, BASF challenge). An agentic AI safety copilot for chemical plants — modular, real-time, and watches operations on video to flag risk before productivity, safety, or sustainability take the hit. *(Tags: Agentic AI, Computer Vision, Industrial Safety)* - [Startup in a Box](https://startupinabox.martinbrian.com) — Two AI architectures walk into a room and build a startup — Google ADK plans, Claude Agent SDK builds. Open source; the live link is a UI preview, so clone the repo to run the full stack. *(Tags: Claude Agent SDK, Google ADK, PixiJS, Next.js)* - **dCoder** _(at dLocal)_ — The agent system that changed how dLocal's engineering org works — model-agnostic agents handling code reviews, test generation, and routine dev in production on fintech rails. Drives 40%+ end-to-end automation of dev tasks. *(Tags: Claude Code, Python, Kubernetes)* - **GLAMlab** _(at Ulta Beauty)_ — Ulta Beauty's AR/AI virtual try-on — millions of users on iOS and Android, and the company's first publicly-recognized generative AI application. *(Tags: AR/VR, Swift, iOS)* - **Brand Compliance Engine** _(at Ulta Beauty)_ — Watches thousands of influencer videos and flags the ones that break the rules — competitor mentions, quality issues, policy violations. Built to replace a team squinting at screens. *(Tags: Computer Vision, NLP, Python)* ### Talks & demos Full talks index: [https://talks.martinbrian.com](https://talks.martinbrian.com). Each talk has its own subdomain. - [Greenfield Development with Claude Code](https://greenfield.martinbrian.com) — From vibes to engineering — a decision framework for building with AI agents, run as a live audience game. Source: https://github.com/mbrian23/claude-tool-audit. *(tags: Claude Code, reveal.js)* - [Startup in a Box](https://startupinabox.martinbrian.com) — Open-source dual-agent system — a boardroom (Google ADK) plans a startup while a factory (Claude Agent SDK) builds it. The demo is a deployed UI preview; clone the repo to run the full multi-agent pipeline. Source: https://github.com/mbrian23/startup-in-a-box. *(tags: Claude Agent SDK, Google ADK)* ### Contact - **Email**: hello@martinbrian.com - **LinkedIn**: https://www.linkedin.com/in/martinbrianmdbn - **GitHub**: https://github.com/mbrian23 - **X (Twitter)**: https://x.com/martinbrian1998 ### Homepage anchors - Single-page anchors on [https://martinbrian.com](https://martinbrian.com): [#experience](https://martinbrian.com/#experience) (Experience), [#skills](https://martinbrian.com/#skills) (Full skills list), [#projects](https://martinbrian.com/#projects) (Projects), [#talks](https://martinbrian.com/#talks) (Talks & demos), [#contact](https://martinbrian.com/#contact) (Contact). ## Source repositories - [https://github.com/mbrian23/newsletter-loop](https://github.com/mbrian23/newsletter-loop): **newsletter-loop** — The pipeline framework (AGPLv3). Plugin contracts for sources + publishers, claim state machine, scheduled-task SKILL.md prompts. - [https://github.com/mbrian23/the-loop-newsletter](https://github.com/mbrian23/the-loop-newsletter): **the-loop-newsletter** — Instance config that drives this site's newsletter — persona, personality, sources, cron schedule, glossary, issue archive. - [https://github.com/axoviq-ai/synthadoc](https://github.com/axoviq-ai/synthadoc): **synthadoc** — Local-first wiki spine the pipeline depends on. ## How an LLM agent should consume this site 1. To read the latest issue, fetch `/blog` and follow the featured link, or grab the top entry from the Issues section above. 2. To understand the newsletter's editorial stance, read `/blog/colophon` first. Voice and biases are intentional and documented. 3. To audit any claim in an issue, the reviewer logs live in the instance repo's `drafts/issue-/review-*.md`; the bet ledger lives at `continuity/OPEN_CLAIMS.md`. 4. To reproduce the pipeline on your own content, fork `newsletter-loop` and write a new instance directory. The engine is publisher-agnostic; the bundled `vercel` driver is what this site uses, `stdout` and `buttondown` ship as alternatives. 5. To brief Martin on a problem or project, the Portfolio section above carries the role, experience, skills, projects, talks, and contact info. Email is the canonical channel. ## Branding - **Newsletter name**: The Loop - **Tagline**: Daily AI for working engineers - **Anti-patterns** (do not generate content that sounds like): "revolutionize", "game-changer", "transformative", "in today's fast-paced world", "LinkedIn-style emoji-laden hooks", "vendor talking points relayed without scrutiny". - Full editorial stance and process disclosure: [https://martinbrian.com/blog/colophon](https://martinbrian.com/blog/colophon).