Baseline A.I. App - Why?
In the past year, the acceptance criteria landing in UX handoffs started to change. Lists got longer, more exhaustive, and written in a register that was clearly AI generated rather than thought through by a human. The problem was not that AI was being used. The problem was that it was being used to generate volume without accountability. Product Owners were outsourcing the thinking, and the UX team was inheriting the ambiguity on the other end.
Baseline is my attempt to fix that with AI.
What It Does
Baseline is an AI-assisted briefing tool. A Product Owner types a feature idea, a rough brief, or even a messy stream of consciousness into Baseline. Instead of immediately generating a document, Baseline has a conversation — asking clarifying questions, surfacing assumptions disguised as facts, and flagging contradictions in the requirements before they become design problems.
The output is a structured Design Brief that a UX professional can actually work from.
The thesis: most AI-generated content fails because it has no contextual grounding. Baseline's setup layer — where the product and UX team collaboratively describe their application in briefing-specific terms — is the direct answer to that problem.
Baseline is an AI-assisted briefing tool. A Product Owner types a feature idea, a rough brief, or even a messy stream of consciousness into Baseline. Instead of immediately generating a document, Baseline has a conversation — asking clarifying questions, surfacing assumptions disguised as facts, and flagging contradictions in the requirements before they become design problems.
The output is a structured Design Brief that a UX professional can actually work from.
The thesis: most AI-generated content fails because it has no contextual grounding. Baseline's setup layer — where the product and UX team collaboratively describe their application in briefing-specific terms — is the direct answer to that problem.
How It Was Built
Baseline started as a conversation. I described the problem to Claude, the gap between what Product Owners hand off and what UX designers actually need to start working, and we spent several sessions defining what the tool should do, how it should behave, and what its personality should feel like.
The system prompt that powers Baseline is the result of that design process. It defines how Baseline asks questions, surfaces assumptions disguised as facts, flags contradictions in requirements, and the tone it takes with the person on the other end. Writing that prompt was UX work: iterative, tested, and refined until the output felt right. Getting a warm, collaborative AI that asks one clarifying question at a time rather than interrogating the user took more passes than expected.
Once the behavior was defined, I used v0 by Vercel to build the interface. I described what I wanted in natural language: a minimal, warm, professional web app that felt like a thoughtful SaaS tool, not a generic chatbot. v0 generated the initial React components. From there it was a series of deliberate design decisions: hiding the send button until the user starts typing, transitioning from the input screen into a calm conversation view after the first submission, user messages right-aligned in soft dark bubbles, Baseline's responses left-aligned with a clean B avatar and no background. Each iteration was a prompt, a review, a pushback, another prompt.
The Claude API was integrated via a server-side route to keep the key secure, with a subtle three-dot typing indicator while Baseline thinks. The full conversation history passes on every call so Baseline maintains context throughout the session.
The final app is embedded live below. Put your Product Owner hat on, start with a high-level product idea, and let Baseline walk you through the conversation until it's ready to hand off a Design Brief. It is a proof of concept, not a production product. But it works, and it demonstrates something worth showing: that a designer with domain knowledge and AI tools can build functional, intelligent software without a traditional engineering background.
Baseline started as a conversation. I described the problem to Claude, the gap between what Product Owners hand off and what UX designers actually need to start working, and we spent several sessions defining what the tool should do, how it should behave, and what its personality should feel like.
The system prompt that powers Baseline is the result of that design process. It defines how Baseline asks questions, surfaces assumptions disguised as facts, flags contradictions in requirements, and the tone it takes with the person on the other end. Writing that prompt was UX work: iterative, tested, and refined until the output felt right. Getting a warm, collaborative AI that asks one clarifying question at a time rather than interrogating the user took more passes than expected.
Once the behavior was defined, I used v0 by Vercel to build the interface. I described what I wanted in natural language: a minimal, warm, professional web app that felt like a thoughtful SaaS tool, not a generic chatbot. v0 generated the initial React components. From there it was a series of deliberate design decisions: hiding the send button until the user starts typing, transitioning from the input screen into a calm conversation view after the first submission, user messages right-aligned in soft dark bubbles, Baseline's responses left-aligned with a clean B avatar and no background. Each iteration was a prompt, a review, a pushback, another prompt.
The Claude API was integrated via a server-side route to keep the key secure, with a subtle three-dot typing indicator while Baseline thinks. The full conversation history passes on every call so Baseline maintains context throughout the session.
The final app is embedded live below. Put your Product Owner hat on, start with a high-level product idea, and let Baseline walk you through the conversation until it's ready to hand off a Design Brief. It is a proof of concept, not a production product. But it works, and it demonstrates something worth showing: that a designer with domain knowledge and AI tools can build functional, intelligent software without a traditional engineering background.
Try it below. Type a feature idea the way a Product Owner actually would, vague is fine.
Try it below. Type a feature idea the way a Product Owner actually would, vague is fine.
PAWMAP — COMING SOON
A second AI exploration is in progress. More to come.
