Industry:
AI-native sustainability tech
Role:
Fullstack designer: concept, design, product strategy, building MVP for closed Beta tests.
Tools & systems used
Figma, Lovable, Supabase
mission.
Designing an AI-native product for regenerative choices
As the sole product designer, I led the strategy, UX, and UI for Archaster AI, a 0-to-1 product aimed at embedding expert-level sustainability guidance into the workflow of professionals.
the challenge.
The illusion of the "all-knowing" AI
Professionals in architecture, design, and procurement are under increasing pressure to make sustainable, regenerative choices. But our research has shown that they often lack the specialized knowledge to evaluate new materials, navigate complex supply chains, or apply circular economy and planetary boundary principles effectively.
The rise of Large Language Models presented a potential solution, but one with a critical flaw: the "Blank canvas problem."
Generic AI chat interfaces fail because:
Users don't know what to ask. The domain is too complex for a non-sustainability-expert to formulate a precise, effective prompt.
The AI lacks context. Without knowing the user's role or intent, the AI provides generic, low-value answers.
This leads to a frustrating loop of vague questions and unhelpful responses, causing high user churn and a failure to demonstrate the AI's true potential.
The goal was to design an AI-native product that solved this cold start problem for both the user and the AI, creating a path to immediate value.
solution.
The persona-to-prompt pipeline
The AI chat activation screen is not a simple welcome page like most other chat screens; it's a strategic engine designed to solve the cold start problem in a single, elegant interaction.
How it works:
Guided onboarding: Instead of an intimidating blank text box, the user is greeted with a clear value proposition and a simple question: "What's your challenge?" This immediately frames the AI as a collaborative partner.
Persona-based context Injection: The "Choose what describes you best" section is the most critical strategic element. When a user self-selects their role (e.g., "Procurement"), two things happen:
For the AI: A hidden context is injected into the system prompt. The AI now knows it's speaking to a procurement professional and can tailor its language, knowledge, and recommendations accordingly.
For the user: The interface dynamically updates to show persona-specific suggested prompts.
"Zero-click" value demonstration: The suggested prompts for "Procurement" ("How do I evaluate suppliers...", "What standards should I use...") are high-value, expert-level questions. This shows the user the AI's power before they even type a word. It guides them directly to their "aha!" moment.
Integrated growth engine: The "[n] beta queries remaining" is a deliberate Product-Led Growth (PLG) mechanic. It meters the product's core value—expert answers—creating a natural and compelling path toward conversion without using frustrating feature gates.
In chat mode, the user can choose between card and document view, catering to different preferences. Both modes have a "copy text" button for simple export.
outcome.
From concept to MVP in 10 days
This design successfully transformed a simple chat interface into a sophisticated user activation and Go-to-Market machine. For native AI products, the most critical design challenge is the strategic design of the first interaction. Success lies in designing the conversation, not just the interface.
Solved the AI cold start: The Persona-to-Prompt pipeline eliminates user anxiety, prevents generic queries, and ensures the AI delivers high-quality, contextual answers from the very first interaction.
Built a GTM & PLG engine: The design is a system for user segmentation, product-led growth, and identifying high-intent B2B leads, providing invaluable data for the business.
ongoing.
Closed Beta
The product is now in closed Beta and user research is being done with the early adopters. Critical user journeys are tracked to inform further product development.
Full case study available on request — email me!
