Own product · AI SaaS · 2024 · architecture case study
Self-Serve Chatbot Platform
URL in, grounded chatbot out — scrape a site into a knowledge base and return an embeddable widget.
- RAG
- Pinecone
- Embeddable widget
- Lead-gen
The problem
A small business wants a chatbot that answers visitor questions and captures leads, but hiring an agency to build one costs money and takes weeks — and most small businesses have no engineer on staff to do it themselves. The gap isn’t ambition, it’s access.
What I built
Give the system a business’s URL and it scrapes the site into a knowledge base, indexes it with retrieval-augmented generation over Pinecone, and returns an embeddable JavaScript snippet the owner drops into their own site. Around the chatbot itself: lead-gen forms to capture visitor contact details, and an analytics dashboard so the owner can see what people are actually asking.
Decisions worth noting
- Grounded in scraped content only. The chatbot answers from what it indexed off the business’s own site — no open-ended generation that can wander into claims the business never made.
- The widget is a guest on someone else’s site. A constraint I’d meet again later: it has to install with one script tag and never break the host page.
- Lead capture is the business model, not the chat. The chat is the hook; the forms and the dashboard are what the owner is actually paying for.
My role
Built solo, own IP, in my freelance era in 2024 — before Alyne. Honest arc: I built this product category twice. This was the first pass; the AI Portal is the second, better iteration of the same idea, with a more mature build behind it. No screenshots exist from this build, so this page is the architecture, not a screenshot tour.