Sage: Evolving Expert System

Sage is an AI expert you build on a specific domain. Unlike generic AI assistants, Sage starts with your knowledge — documents, URLs, or text you upload — and gets better at answering questions in that domain the more it's used.

How Sage is built

You create a Sage by giving it:

• A name and domain (e.g., "Dr. Chen — pediatric nutrition")
• Knowledge sources — PDFs, URLs, or text that represent what the Sage should know

The AI extracts the knowledge, organises it into a structured knowledge base, and the Sage is ready to answer questions from anyone you share it with. Initial setup takes minutes once you have the source material.

How users interact with Sage

Once built, the Sage can be exposed publicly — anyone can chat with it, ask questions in the domain, and get answers grounded in the knowledge you provided. This is the "deployment" of the expert: you've built the Sage, now it serves questions.

Public chats with the Sage are recorded. You (as the Sage's owner) can review what people are asking and how the Sage is answering.

How Sage improves over time

Sage isn't static. After enough public chats accumulate, you can analyse recent conversations to find knowledge gaps — questions the Sage answered poorly or couldn't answer at all. Gaps are flagged with severity (critical, important, nice-to-have) and linked back to the source conversations.

You then run a supplementary interview with the Sage to fill those gaps. The interview is a focused conversation where the AI asks you questions specifically designed to cover what's missing. At the end of the interview, the new knowledge is added to the Sage's working memory.

Periodically, you integrate working memory into the Sage's stable core knowledge. Each integration creates a new version of the Sage, so the evolution is traceable.

The improvement loop

The cycle is:

1. Build — upload sources, Sage learns the initial knowledge.
2. Deploy — share the Sage publicly, users chat with it.
3. Analyse — review recent chats, identify knowledge gaps.
4. Interview — fill gaps through focused Q&A with the Sage.
5. Integrate — fold the new knowledge into the Sage's stable base.

Repeat this loop. Each iteration makes the Sage more capable in its domain.

Why this design

The loop exists because real expertise isn't just documents — it's the questions real users ask that you didn't anticipate. By analysing conversations, you discover what the Sage is missing that no source document would have told you.

The supplementary interview is the mechanism for capturing your tacit knowledge (the things you know but never wrote down). The integration step keeps the stable knowledge base coherent instead of fragmenting it across many incremental edits.

Limits

• Sage is only as good as the knowledge you give it. If your sources are shallow, the Sage will be too.
• Knowledge gap analysis requires enough conversations to be meaningful — typically 20 or more recent chats.
• Sage doesn't learn on its own. The improvement loop requires you (the owner) to actively review gaps and run interviews.
• Sage answers are grounded in its knowledge base. It won't fabricate information outside the sources you provided, but it also can't answer questions it has no basis for.
• Sage is domain-specific. A pediatric-nutrition Sage won't suddenly become a finance Sage — the domain is fixed at creation.

When Sage is the right tool

Use Sage when:

• You have a body of knowledge you want to deploy at scale (customer education, internal Q&A, expert consultation).
• You expect users to ask questions your standard docs can't cover well.
• You want the system to surface what it doesn't know, so you can systematically fill gaps.
• You have the time to run the improvement loop — Sage isn't fire-and-forget.

Skip Sage when:

• You just need a FAQ page — write static docs instead.
• Your domain changes too fast for stable knowledge sources (live news, trading signals).
• You don't have the bandwidth to review conversations and run interviews.

Practical example

You run a wine import business. You build a Sage on Italian wine regions, grape varieties, food pairings, and your product catalog. Customers visit the Sage page, ask about pairings for specific dishes, get recommendations grounded in your catalog. After 100 conversations, you run gap analysis: customers ask about organic certification, but your sources don't cover it. You interview yourself about your organic sourcing policy, integrate it, and the Sage now answers those questions too.