Locus AI
Overview
Building an AI decision memory layer from scratch, as the founding designer.
Timeline
06.2026-Present
Role
Lead Product Designer
Team
Lead Product Designer (me!)
Product Manager
Product Designer *2
Tool
Figma
Status
MVP1 Launched
Locus AI
Overview
Building an AI decision memory layer from scratch, as the founding designer.
Timeline
06.2026-Present
Role
Lead Product Designer
Tool
Status
MVP1 Launched
Team
Product Designer (me!)
Product Manager
Product Designer *2
Background
Teams make decisions everywhere — Slack threads, Notion docs, email chains, video calls. Then nobody can find them.
Locus AI captures and surfaces your team's decisions automatically, so the knowledge stays even when the conversation moves on.
Problem
I joined as the founding designer and led the design effort end-to-end. No UI, no design system, no defined flows. To cover the ground of a small design team at startup speed, I built an AI-assisted pipeline into every stage of the project.
Solution
Four features that turn scattered team conversations into searchable, structured knowledge.
Dashboard: Every decision your team has made — captured automatically, surfaced at a glance.
Memory Explorer: Search across every conversation, channel, and tool — by topic, person, or date.
Memory Timeline: A chronological view of how your team's decisions evolved over time.
4. Team Pulse: Who's making decisions, where conversations are happening, and where knowledge might be siloed.
So…How did i get there…?
So…How did i get there…?
Research
Competitive analysis
Internal Research
Internal research shows that the documentation webiste is not efficient enough in solving user’s operation problems, especially when the issues are trivial.
Insights
💡
User should have quick access to frequently asked questions.
💡
User should be able to solve problems in a various range of types without contacting technical support.
Research
Competitive analysis
Internal Research
Internal research shows that the documentation webiste is not efficient enough in solving user’s operation problems, especially when the issues are trivial.
Insights
💡
User should have quick access to frequently asked questions.
💡
User should be able to solve problems in a various range of types without contacting technical support.
AI-Assited Workflow
I designed a multi-tool AI pipeline that let me move from PRD to shipped design system in weeks — not months. AI handled the repetitive and generative work. I made every design decision.
Understand
Claude, ChatGPT
Used Claude and ChatGPT to digest the product requirements and map out the information architecture. A working IA structure within hours, not days.
Explore
Google Stitch
Fed the IA into Google Stitch to rapidly generate design directions, UI styles, and layout variations. Dozens of visual options to react to instead of a blank canvas.
Systematize
Claude, Claude Code, Figma MCP
Claude read the Stitch-generated designs and extracted UI variables — colors, spacing, typography — formatted for direct Figma import. Then Claude Code with Figma MCP built and audited a 100+ component design system for consistency and usability.
Prototype
Claude
Built a unified HTML prototype with Claude to validate the full user flow before engineering started. The team experienced the entire product as a clickable prototype — not static screens.
Refine
Claude
Used Figma's AI agent to audit high-fidelity designs against the system — catching inconsistencies, accessibility issues, and component misuse before handoff.
Design Principles
🧐
Prioritize Findability Over Volume
🚀
Enable Fast Switching Between Help Sources
🧩
Adapt to User Scenarios with Flexible UI
Design Principles
🧐
Prioritize Findability Over Volume
🚀
Enable Fast Switching Between Help Sources
🧐
Prioritize Findability Over Volume
Takeaways
🔍Find problems
Instead of waiting to be informed a problem, a qualified designer should be able to spot potential problems and plan ahead.
🧩Think in System, Not Screens
I learned to zoom out from the screen I was designing and ask: how does this decision affect the whole product? What would the product be like in the future?
Locus AI
Overview
Building an AI decision memory layer from scratch, as the founding designer.
Timeline
06.2026-Present
Role
Lead Product Designer
Team
Lead Product Designer (me!)
Product Manager
Product Designer *2
Tool
Figma
Status
MVP1 Launched
Background
Teams make decisions everywhere — Slack threads, Notion docs, email chains, video calls. Then nobody can find them.
Locus AI captures and surfaces your team's decisions automatically, so the knowledge stays even when the conversation moves on.
Problem
I joined as the founding designer and led the design effort end-to-end. No UI, no design system, no defined flows. To cover the ground of a small design team at startup speed, I built an AI-assisted pipeline into every stage of the project.
Solution
Four features that turn scattered team conversations into searchable, structured knowledge.
Dashboard: Every decision your team has made — captured automatically, surfaced at a glance.
Memory Explorer: Search across every conversation, channel, and tool — by topic, person, or date.
Memory Timeline: A chronological view of how your team's decisions evolved over time.
4. Team Pulse: Who's making decisions, where conversations are happening, and where knowledge might be siloed.
So…How did i get there…?
Research
Competitive analysis
Internal Research
Internal research shows that the documentation webiste is not efficient enough in solving user’s operation problems, especially when the issues are trivial.
Insights
💡
User should have quick access to frequently asked questions.
💡
User should be able to solve problems in a various range of types without contacting technical support.
AI-Assited Workflow
I designed a multi-tool AI pipeline that let me move from PRD to shipped design system in weeks — not months. AI handled the repetitive and generative work. I made every design decision.
Understand
Claude, ChatGPT
Used Claude and ChatGPT to digest the product requirements and map out the information architecture. A working IA structure within hours, not days.
Explore
Google Stitch
Fed the IA into Google Stitch to rapidly generate design directions, UI styles, and layout variations. Dozens of visual options to react to instead of a blank canvas.
Systematize
Claude, Claude Code, Figma MCP
Claude read the Stitch-generated designs and extracted UI variables — colors, spacing, typography — formatted for direct Figma import. Then Claude Code with Figma MCP built and audited a 100+ component design system for consistency and usability.
Prototype
Claude
Built a unified HTML prototype with Claude to validate the full user flow before engineering started. The team experienced the entire product as a clickable prototype — not static screens.
Refine
Claude
Used Figma's AI agent to audit high-fidelity designs against the system — catching inconsistencies, accessibility issues, and component misuse before handoff.
Design Principles
🧐
Prioritize Findability Over Volume
🚀
Enable Fast Switching Between Help Sources
🧩
Adapt to User Scenarios with Flexible UI
Takeaways
🔍Find problems
Instead of waiting to be informed a problem, a qualified designer should be able to spot potential problems and plan ahead.
🧩Think in System, Not Screens
I learned to zoom out from the screen I was designing and ask: how does this decision affect the whole product? What would the product be like in the future?
