> ## Content Index > Fetch the complete content index at: https://intelligence.artificialrealitymedia.com/llms.txt > Use this file to discover other available public pages before exploring further. # How a non-developer built a fifteen-skill assessment system in Claude - URL: https://intelligence.artificialrealitymedia.com/non-developer-fifteen-skill-claude-build/ - Published: 2026-09-02T12:01:00.000Z - Updated: 2026-09-02T12:01:00.000Z - Description: Dr. Nikki Siso started with one Claude skill, hit a depth limit, and split it into eight, then fifteen. Nobody sold her that architecture. - Author: Galina Fendikevich - Tags: Restaurants, #For AI Companies, #What's working, #Import 2026-09-03 05:09 Dr. Nikki Siso is not a developer. The certified holistic health practitioner and founder of Conscious Kitchen, a raw, organic, plant-based, gluten-free restaurant in Westlake Hills outside Austin, built her total toxic load assessment as one skill in [Claude](https://claude.ai/?ref=intelligence.artificialrealitymedia.com), found it too deep to run efficiently, and split it into eight skills, then fifteen. Nobody sold her that architecture. She hit a limit and worked it out, which is the part worth reading if you build for people like her. ## What did she actually build in Claude? A multi-skill client assessment, separated by domain, for the toxic load work in her holistic health practice. > "For my total toxic load assessment, I built out the skills in Claude. It started off as one skill and then it became way too deep. And I broke it out into eight different skills, now 15 — specific to what are the toxins in the home, what are the toxins in the food, what are the toxic thoughts that lead to dis-ease in the body." ## Why did one skill have to become fifteen? Depth, in her own description. Each subject area carried its own body of research, and a single container could not work through that much material efficiently. > "Each one of those had a bank of wisdom and knowledge and peer-reviewed studies that required a depth that one skill couldn't read through that much ... efficiently." Note the shape of that decision. Siso did not decompose the system because a best-practice guide told her to. She built the obvious thing, watched it degrade as the corpus grew, and partitioned by subject matter — home, food, thought — because that is how her own practice is organised. The taxonomy came from the domain, not from the tool. One to eight to fifteen is an architecture lesson arrived at by hitting a wall, not by reading documentation. If your product expects a non-technical buyer to design that partition up front, this is the failure mode you are shipping. ## How did the one skill she did not build reach her? Through a friend, not a marketplace. Her market research runs on a skill somebody she knows made. > "I have a skill my friend developed called Market Sauce, and it is a deep dive into everything you need to know about your target audience. Who they actually are, what the size, how they want to be talked to, what are their biggest concerns." Her description of what it amounts to: "He's created a kind of chief marketing AI officer for you. So you put your concept in and then it prints this full document for you." Two signals there. The distribution path was a person she knows. And the unit she values is a finished document she can act on, not a chat session she has to steer. ## What did that research actually change? It killed the assumption the restaurant would have died on. > "Who's our target audience? Are they just raw vegans? Well, there are like four in Austin. We'd be out of business very quickly." What surfaced instead was a set of distinct groups — people avoiding gluten for celiac disease, people managing health conditions, people choosing plant-based for ethical reasons — each of whom needs to be spoken to differently. For a small restaurant with one menu, that segmentation is the difference between a business and four customers. ## How does she get value out of a model? By refusing the first answer. On understanding how food preservatives are designed: > "I dug into this on [ChatGPT](https://chatgpt.com/?ref=intelligence.artificialrealitymedia.com) actually. I was like: okay, tell me exactly how preservatives are designed. And it gave me this superficial answer. I was like: no, no, no, I want the science." The second answer was the useful one, and it became part of how she explains the restaurant's premise to customers. Treating the first response as a draft and asking for the mechanism separates operators who get value from these tools from those who do not. ## Which tools were named on the record? - **Claude** — the total toxic load assessment, one skill split into eight and then fifteen. Worked, after she restructured it herself. - **Market Sauce** — a skill built by a friend, used for target-audience research. Worked. - **ChatGPT** — research into preservative design. Worked on the second prompt, not the first. Tools named by an operator on the record. Inclusion is reporting, not endorsement. ## What should a builder take from this? My read: Siso is the customer most AI products claim to serve and very few are shaped for. No technical background, no consultant budget, high tolerance for iteration, and enough domain expertise to specify a system precisely. Her fifteen-part architecture exists only because she kept going after the first version got worse. The gap she exposes is not capability. It is the moment the single-container version stops working. Nothing warned her it was coming, nothing proposed the split, and nothing carried the structure for her once she made it. If your honest answer to "what happens when the knowledge base outgrows one container" is that the user figures it out, you have handed your least technical buyer your hardest problem. Two things to change this quarter. Instrument the point where a customer's own corpus starts degrading their results, and tell them before they feel it. And look at how Market Sauce reached her: built by a friend, delivering a document rather than a conversation. What is your product's version of that? ## Where this comes from S1E6: I trusted AI to build my restaurant, it cost me $30,000 — the full interview with Dr. Nikki Siso. Listen or watch: [YouTube](https://www.youtube.com/watch?v=7LF4Ro8Hg6M&ref=intelligence.artificialrealitymedia.com), [Spotify](https://open.spotify.com/episode/5TwX8rLnDKkne9REQjFCEi?ref=intelligence.artificialrealitymedia.com) or [Apple Podcasts](https://podcasts.apple.com/us/podcast/i-trusted-ai-to-build-my-restaurant-it-cost-me-%2430-000/id6780699025?i=1000777268211&ref=intelligence.artificialrealitymedia.com).