Pierre Rognion

Forward-deployed delivery for one of France's largest freelance marketplaces: business problem → system design → build → production → adoption, working with Sales, Product and Engineering. Systems designed, coded and shipped to production, in daily use: - A hybrid search and ranking pipeline for talent matching, built as an eval harness that had to beat the existing production baseline before it was allowed to ship. - A human-in-the-loop decision pipeline with schema-validated multi-state verdicts and rule-based routing of edge cases to a human. - Natural-language access to a large production database: read-only by governance choice, with business rules wired into mandatory query templates. - Internal MCP servers over the company's tooling, plus an MCP server generating design-system-aligned front-end mockups. - A marketplace of modular, recombinable Claude skills packaging Sales know-how. - Agentic external sourcing and enrichment beyond the internal base. Also retired several early systems once the incumbent SaaS shipped equivalent. Knowing when to stop building is part of the job! Stack: TypeScript, Python, React, Node, PostgreSQL, MCP, Claude, OpenAI, Vercel, PostHog.

My own company (SASU), through which I run my missions and my own builds. Custom agents, internal prototypes, and hands-on training programmes to accelerate AI adoption. Built InsightX, an LLM tool that retrieves and analyses SEC 10-K filings for interactive exploration, and a multi-LLM chatbot with web search. I've trained 1,000+ professionals in applied GenAI at Chanel, Stellantis, Bouygues Telecom, Rakuten, Generali, Institut Pasteur, and many others...

Hands-on GenAI programmes for Product and Tech teams: prompt design, AI-assisted ideation, workflow automation.