When I entered the team, the product had just released its first MVP. My role began with understanding its real-world performance including reviewing analytic data, observing user behaviors, and interviewing early adopters.

I. Research Process

1.1 — MVP Critique

When I joined, one of my first tasks was to examine how users were interacting with the MVP. Although the product successfully aggregated thousands of AI tools, the experience revealed a deeper issue: users didn’t know how to navigate or express what they needed.

Web MVP

Mobile MVP

HMW create an intuitive, intent-aware discovery experience that helps users start easily, refine confidently, and evaluate tools with minimal effort?

1.2 — Understand the Industry

2024 marked a turning point in how people interacted with AI. As large language models(LLMs) became mainstream, users grew more comfortable expressing needs in natural language—but far less comfortable navigating categories, keywords, or technical terminology.

Most users knew what they wanted to achieve, but not what kind of AI tool, GPT, or API could help them do it. Meanwhile, the ecosystem expanded faster than user literacy: thousands of tools, overlapping features, inconsistent naming, and unclear differentiation.

1.3 — AI Product Design Principles

By aligning model strengths with market needs, I shaped the value proposition that became the north star of every design decision:

01 — Align With User Nature, Not How Tools Are Categorized

Users don’t think in keywords or categories—they think in goals, tasks, and situations.Our interface starts with natural language input and structures it into meaningful intent dimensions (task, goal, constraints, context).

02 — Make the System’s Understanding Visible

Every step of the process—intent extraction, categorization, reasoning—should be legible to the user.We expose the system’s interpretation of their text so users can validate and correct it.

03 — Reduce Cognitive Load by Guiding Momentum

Instead of confronting users with dozens of filters or choices, the system guides users forward with light-touch refinement steps, recommended categories, and curated results.

04 — Design for Ambiguity, Not Perfection

User inputs are often vague, incomplete, or incorrect.The system must handle ambiguity gracefully—asking clarifying questions, inferring context, and offering suggestions.

05 — Curate Results With Contextual Meaning

Results (AI tools, GPTs, APIs) should not be displayed as a flat list. They should be framed by:why they match the user’s intentwhat makes them differenthow they solve the user’s goal

06 — Maintain Structural Integrity Across Platforms

Discovery should feel consistent whether users are on desktop or mobile.Layouts, refinement steps, and intent breakdowns follow a shared logic—even as their presentation adapts.

1.4 — User Scenario Analysis

While defining the value story, I also Interviewed a few potential users recruited from huggingface and LinkedIn—mainly engineers familiar with evaluating new models.

From these scenarios, it became clear that the product needed to support multiple entry points: exploration without a clear goal, task-driven discovery, and feature-specific queries.

I translated these needs into an information architecture that balanced flexibility with clarity, ensuring AI-driven suggestions were at the heart of the experience.

1.5 — Information Architecture

While defining the value story, I also Interviewed a few potential users recruited from huggingface and LinkedIn—mainly engineers familiar with evaluating new models.