Sean Carroll

BikeWise (Exited) · Paris, France

Co-founded BikeWise, which exited in March 2026, and built SheldonAI from zero — an enterprise-grade, agentic RAG platform for the cycling industry. Designed ingestion and embedding pipelines over 2.5M+ product listings, CLIP and text embeddings for hybrid retrieval, and LangGraph orchestration for query-time reasoning so operators get semantic search, recommendations, and inventory intelligence instead of brittle keyword filters. Owned the full lifecycle solo through EIT Urban Mobility Accelerator recognition, then landed and supported enterprise customers via customer engineering and white-glove API integration.

• ·Production RAG at catalog scale: deduplication, chunking strategies, vector indexes, reranking, and eval loops tuned against real merchandiser feedback — not demo retrieval
• ·Shipped APIs and dashboards operators actually run in production; paired FastAPI services with a Next.js/Tailwind front end, Supabase/Postgres, Redis, and Gemini where multimodal reasoning helped close the loop

Freelance · Remote

Fractional CTO and hands-on technical partner for AI-forward startups from idea stage through Series B-scale pressure. Embed with founders and product teams to turn ambiguous requirements into shippable systems: technical discovery, architecture, build-or-buy decisions, vendor and model selection, and the people/process work (hiring rubrics, code review culture, CI/CD) that keeps velocity honest after the first demo.

• ·Portfolio spans $0–$25M ARR across home health care, real estate, e-commerce, and social — same playbook: map the constraint (compliance, data quality, latency, cost), design the smallest production slice, and ship with observability from day one
• ·Deliver LangGraph (and adjacent) agent workflows, RAG over messy domain corpora, eval harnesses, and LLM-backed APIs that respect tenancy, PII, and each client's infra maturity — usually Python/FastAPI services paired with Next.js or existing app surfaces
• ·Act as the single thread between execs, domain experts, and engineers: roadmap facilitation, sprint-level prioritization, interview loops, and crisp written specs so teams do not stall on alignment

Partnering with startups and growth-stage teams for fractional CTO and strategic technical leadership. Open to selective new engagements.

Madelon · New York, NY, USA

Startup CTO building REDtech — a GIS/ML platform that fuses parcel geometry, zoning rules, and proprietary underwriting logic so acquisition teams could compare sites in hours instead of weeks. Worked directly with national real estate developers to shape the product around how their acquisition teams actually underwrite. Owned product direction, data engineering for spatial joins and ETL from municipal sources, and the React/JavaScript application surface analysts used daily. Scaled the team from zero to four engineers on a $250K annual budget with nearshore partners, clear architecture reviews, and predictable release cadence across multiple active markets.

• ·PostGIS-first modeling, Mapbox-backed visualization, and custom ML heuristics for lot-yield and constraint scoring — the same systems-thinking later applied to AI feature work, but here tuned for capital markets rigor
• ·Hiring, budgeting, and stakeholder management in a regulated, document-heavy domain; learned how to ship when the source of truth is PDFs, GIS layers, and Excel, not a clean SaaS schema

ADK Group · Boston, MA, USA

Delivered full-stack Drupal and React engagements for agency clients: custom modules, editorial workflows, headless or hybrid integrations, and front-end performance work where Core Web Vitals and perceived speed were part of the contract. Partnered with designers and PMs in Jira-driven sprints; habits from this era (clear tickets, incremental releases, client demos) still shape how larger AI projects get governed.