Waterside App

Waterside Group runs 25+ businesses across marine, hospitality, real estate, construction, and recreation, and the connective work ran the way it always had: scattered email threads, shared spreadsheets, and answers that lived in someone's inbox. Requesting marketing work, tracking a ticket, or generating a document meant chasing people instead of using a system.

I built the replacement hands-on in agentic AI coding environments (Claude Code and the Claude API): a central nervous system for the operating group on Next.js and Supabase, with a ticketing and marketing hub, company directory, calendar, document generation, and an AI assistant layer, organized as modular hubs for Marketing tickets, Companies, Accounting, Construction, HR, and Blockers and Manager Reports.

AI runs through the product, not beside it: Anthropic Claude powers the assistant via the Vercel AI SDK, Voyage embeddings drive RAG over company knowledge, headless browser automation with the Steel SDK and Playwright handles data capture, and Word, Excel, and PDF deliverables generate server side. Everything sits on a shared multi-app Supabase database with Postgres, Auth, Storage, RLS, and Edge Functions.

Staff ask questions in plain language and get answers drawn from the group's own material: the company-wide RAG assistant is one of three production LLM assistants I have shipped for daily use by non-engineering staff.

Staff across the portfolio now request marketing work, track tickets, generate documents, and get answers from a RAG assistant built on company knowledge. The manual workflows I had firsthand exposure to earlier in my time at Waterside are gone.