Challenge

A legal publishing house wanted to do more than use AI — it wanted to offer an AI product of its own: a chatbot built on the publisher’s specialist content that gives customers something generic assistants cannot. The questions at the outset were the classic ones of any product venture: Which use case carries the product? What are the USPs that set it apart from a standard chatbot? And how do you get reliably from an idea to a product that holds up in the market?

Approach

I accompanied the venture end to end — from ideation to go-live:

  • Ideation: developing, evaluating, and sharpening use cases; working out the customer-specific USPs that lift the chatbot above a generic assistant.
  • Project management: requirements, prioritization, steering the development, alignment with business departments and executive management — running the venture as a product project, not a technology experiment.
  • Quality assurance with LLM-as-a-judge: instead of spot-checking answers by hand, a second language model evaluates the bot’s answers systematically against defined criteria — such as factual correctness, grounding in the publisher’s content, and tone. Every iteration becomes measurable, regressions surface immediately, and quality becomes a repeatable process rather than a gut feeling.
  • Path to market: everything a product launch demands — clarifying responsibilities, operations, legal and data-protection questions, preparing communication and support — through to go-live.

Outcome

The chatbot is fully developed and on the market. The publisher has an AI product of its own with clear USPs — and, with LLM-as-a-judge, a quality process that carries beyond the launch: every further development is checked against the same criteria before it reaches customers.