Project Overview

Kakapo turns the information a business already has into instant answers for the people asking. It reads a workspace's own pages, documents, and questions, then answers from that content rather than guessing. Routine questions get handled the moment they are asked, and the support team keeps its time for the conversations that need a person.

Challenges

Chatbots tend to fail in the same ways, so the product had to:

• Answer from the business's real content instead of producing confident generic replies.
• Keep every workspace's knowledge isolated, with retrieval that never crosses a bot or account boundary.
• Go live on an existing website without a rebuild.
• Make answer quality visible, so an owner can see what was asked and how well it was handled.

Our Solution

We built a retrieval-grounded assistant that a non-technical operator can train and ship:

• Knowledge Ingestion: Website URLs, uploaded files, plain text, and question and answer pairs, chunked and embedded for retrieval.
• Grounded Retrieval: Hybrid vector and keyword search scoped to a single bot, with the model writing the reply from what it retrieves.
• Delivery Anywhere: A public share page, an iframe embed, and a floating chat script that drops into any site.
• Operator Control: Branding, welcome message, suggested prompts, allowed domains, and conversation logs with confidence scores, all managed from MindFull Backoffice.
• Observability: Response caching and end-to-end tracing, so answers stay fast and their quality stays measurable.

Results

• Answers Worth Trusting: Replies come from the workspace's own content, and the logs show which questions the bot handled well.
• Less Repetitive Work: Common questions are answered before they reach a person, at any hour.
• Included, Not Sold Separately: Kakapo is the AI capability of MindFull Backoffice, part of every workspace subscription rather than another tool to buy.