Context
I’ve been exploring the concept of a new app designed to strengthen human connections in an increasingly digital world. With the rapid development of AI and drawing on my background in front-end development from university, I saw this as the right moment to begin building the idea. This project is still a work in progress, and I plan to continue evolving it over time.
This page shares a snapshot of what I’ve learned so far while developing the concept using AI-powered tools such as Figma Make, Cursor, and Claude Code. Throughout the process, I’ve been experimenting with ways to reduce “AI slop” by grounding the work in foundational UX and product methodologies, focusing on usability principles and strong value propositions.
The visual design is subject to change, as my current focus is on improving functionality and learning how to better prompt and guide AI tools to achieve the desired usability and product experience.
We live in an increasingly digital world where people scroll instead of speak. In public spaces, strangers stand side by side but rarely interact. Modern dating and social apps prioritize appearance, endless options, and rapid swiping, which often leads to shallow engagement and difficulty forming lasting relationships.
DeepMeet is an app designed to help people build deeper, more meaningful connections, not exclusively for dating, but for human connection more broadly.
The core principles:
Match based on shared interests and location (music, hobbies, values)
Limit users to only 2 active matches at a time. DeepMeet intentionally reduces cognitive overload to increase conversational quality.
Start conversations anonymously
Gradually unlock identity as trust builds
Encourage real conversation and in-person connections over endless browsing
Step 1: Full PRD Creation
I started by writing a complete Product Requirements Document including:
User personas
Core user flows
Matching logic
Anonymity unlock mechanics
Trust-building milestones
Edge cases and moderation considerations
This helped clarify the logic before touching UI.
Step 2: Fed the PRD into Figma Make
Generated initial interface explorations
Experimented with layouts and flows
Rapidly tested interaction ideas
AI accelerates exploration, but it does not replace intentional system design.
Step 3: Feedback & Iteration
I gathered feedback from:
Engineers
Friends
Potential users
Step 4: Code Prototyping (Cursor)
Attached the PRD and Figma UI screenshots to Cursor
Attempted functional implementation
Began debugging
Observed translation gaps between design and code
AI Workflow Observations
Quality Gaps Exist (AI Slop)
In Figma Make:
A small change can unintentionally affect other elements
Generated layouts sometimes break hierarchy
Design system rules aren’t consistently enforced
In Code:
AI may not fully interpret design systems
State logic can diverge from UX intent
Edge cases require heavy human review
Design Systems Must Be Explicit
If a design system is not clearly structured:
AI improvises
Communication breaks down
Translation errors increase
Markdown files and structured documentation help, but interpretation gaps still occur.
Human Oversight is Non-Negotiable
AI accelerates:
Drafting
Layout generation
Code scaffolding
But it does not replace:
Judgment
System alignment
UX consistency
Ethical design decisions
Learnings
Token optimization isn't just about cost, it's about getting better, more consistent output.
AI speeds up iteration but increases the need for quality governance.
Translation between design and code remains fragile.
Documentation clarity directly impacts AI output quality.
Intentional constraints produce better human experiences.
AI works best when paired with structured thinking and strong human review.
Next Steps
Continuing to debug and tighten the prototype, using what I've learned about prompt structure to reduce iteration cycles.
Building a more explicit design token system (spacing, color, type scales as structured data) to give AI tools less room to improvise.
Taking Anthropic's recent AI + product design courses to deepen technical fluency.
Exploring whether the PRD-as-AI-input workflow can be systematized into a repeatable framework for future projects.