Pascal L.

I build AI agents and automations that work inside a real business: Claude and other LLMs connected to your tools through MCP servers, APIs, webhooks and n8n, with a human approval step on the actions that matter. I also build and fix the software around them: full-stack TypeScript with Next.js and React, Shopify storefronts and WordPress sites. I'm a hands-on systems architect with 20+ years across software delivery, infrastructure and technical rescue. What I take on: AI agents, Claude and MCP - Agents and MCP tool gateways that reach your systems with scoped permissions, audit records and approval before risky actions - Claude Code and coding-agent workflows where a deterministic check and independent review run before anything lands - Retrieval, agent memory and evaluation for assistants that answer from your own data Automation and integrations - n8n, webhook and API workflows for intake, routing, CRM updates, notifications and reporting - Repairing brittle flows: duplicate sends, silent failures, missing retries and unclear ownership Full-stack builds and fixes - Next.js, React, TypeScript and Node.js features, APIs and PostgreSQL data models - Review and hardening of AI-built apps before launch: auth, data access, error handling and deployment - Also: Python, Elixir/Phoenix, PostgreSQL, Docker and Linux Mobile apps (Flutter, Android) - Offline-first Flutter apps with on-device encryption, Google Sign-In and Gmail integration - Automated unit, widget and integration tests, and release builds for Google Play Shopify and WordPress - Theme, product-page and cart fixes; plugin conflicts, speed, security warnings and broken updates - Small, reversible changes, checked on the live site and documented for the owner Technical review and fractional CTO - Architecture and code reviews, go-live checks, Docker/Linux/CI deployment recovery, and a written plan your team can execute Work you can look at in my portfolio: - JarvisMCP: a governed gateway that connects AI agents to business tools. My role covered systems architecture and implementation across capability discovery, bounded execution, authorization, auditability and delivery verification. The implementation is private. - Healthy Glow: a branded Shopify storefront covering theme customization, merchandising, the cart and checkout journey, and launch QA. - Ladie One: a delivered Next.js and TypeScript artist portfolio with editorial galleries. - A reusable WordPress platform where I led the systems architecture and technical direction; the October 2026 private-repository review documented 20 modules and 34 releases. Apps I led, live on Google Play: - CallCue: plans calls around holidays and client follow-ups, with a call queue, outcomes and notes, analytics with CSV export, and contact data encrypted on the device (SQLCipher). Fully offline, with unit, widget and integration tests. - ZeroInbox: turns Gmail cleanup into a game: Google Sign-In, emails grouped by sender, swipe actions with undo, bulk unsubscribe, points and badges, with emails processed on the device. My own delivery workflow is agentic too: coding agents work in isolated branches, a deterministic gate and cross-vendor reviewers check every change, and a person decides what ships. How I work: inspect first, agree the acceptance check, ship the smallest useful change, verify it in your real environment, and leave a handoff your team can run. Prototypes are labeled as prototypes, and client work stays private. Good first engagements: a fixed-price fix or audit, a scoped agent or automation build, or a paid architecture review before you commit to a larger build. Credentials: MIT IDSS Data Science and Machine Learning (June 2025); Anthropic AI Fluency: Frameworks & Foundations.