Portland Old Port / Client work / 2025
A production AI assistant for Portland's Old Port
A retrieval-augmented assistant that answers visitor questions from the district's own content.
The problem
portlandoldport.com covers hundreds of shops, restaurants and events. Visitors arrive with specific questions, like where to eat near the ferry or what's open on Sunday, and a site search can't answer them.
What we did
Working with a Maine SEO agency, we built a retrieval-augmented assistant over the site's content. It uses hybrid search in Supabase with pgvector, so it matches both exact business names and the meaning of a question.
Voyage embeddings handle retrieval and Claude Sonnet writes the answers, grounded in what the site actually says rather than what a model guesses.
The result
The assistant is in production on portlandoldport.com, answering visitor questions from the district's own listings.
Need this?
- Agentic AI & RAG builds
We design and ship AI assistants, agents and MCP servers that run in production, not just in a demo.
Related work
- AEOifyOur product
Building an AEO platform that brands pay for
A production AEO SaaS now used across Conair Corporation brands and by Gordini USA.
AEO & e-commerce growthAgentic AI & RAG builds
- InfillOSOur product
Finding infill lots that pencil, before anyone else does
A feasibility engine covering 132,562 Durham, NC parcels, with a founding developer who is also its investor.
Agentic AI & RAG builds
- Maine residents and boatersOpen source
PublicMCP: public services as a conversation
An open-source MCP server for Maine burn permits, fire danger, tides and marine forecasts.
Agentic AI & RAG builds