befriend: Building an AI-Powered Relationship Assistant from 0-to-1
A conceptual design project created for academic purposes.
befriend is an AI-powered relationship assistant that applies large language models and contextual memory to interpersonal understanding. The goal was to explore how AI can observe patterns across conversations and experiences, helping people reflect on emotional dynamics and communication over time.
Role
Product Designer
Tools
Figma, Adobe CC, Miro, Lovable, GPT-4, Gemini
Timeline
October – December 2025
Skills
Human–AI Interaction, Conversational UX, Prompt Design, Behavioral Pattern Analysis
This case study walks through the end-to-end process using the Double Diamond framework, from discovery and definition to development and delivery.
Introduction
Modern relationships are filled with moments that are emotionally complex but hard to articulate. While people turn to friends, notes apps, or the internet for clarity, these options aren’t always available—or unbiased. befriend explores how an AI-powered assistant can help people decode confusing interactions, reflect on emotional patterns, and support their emotional lives around the clock.
Design Question
How might we use AI to help people navigate their real-life relationships?
To answer this, I analyzed how people seek clarity and emotional support in their relationships, identifying gaps in existing tools. Beyond interviews, I experimented with AI-generated personas to simulate real-world emotional contexts and recurring relationship patterns, expanding the scope of traditional UX research.
Research
User Interviews
To better understand how people seek clarity and reassurance in modern relationships, I conducted 5 semi-structured interviews with users of dating apps who already integrate AI into their communication workflows. Participants described using AI to draft replies, refine tone, and navigate ambiguous interactions.
The research aimed to surface recurring emotional challenges, decision-making behaviors, and trust considerations when relying on AI for interpersonal communication. Findings highlighted gaps in current tools, particularly around long-term context, emotional continuity, and reflective support.
AI Persona – Esther
To extend findings from user interviews, I created an AI persona named Esther, modeled as a college student actively navigating dating apps, social dynamics, and emotionally ambiguous interactions. Key insights, behaviors, and pain points uncovered during research were used as inputs to shape Esther’s perspective and responses.
By grounding Esther in real user data, the persona functioned as a research synthesis tool rather than a hypothetical profile. Esther was used to respond to scenarios, questions, and edge cases, allowing me to test assumptions, explore emotional patterns, and validate how users might seek clarity, reassurance, and guidance through AI-supported reflection.
Ideation
To move quickly from concepts to form, I used a hybrid workflow that paired human judgment with AI-generated output. Tools like Gemini, Lovable, and Figma Make were used to generate early sketches, layout ideas, and interaction patterns, which I then refined manually in Figma. This approach allowed me to test multiple directions in parallel, evaluate trade-offs, and iterate faster than traditional sketching alone, while still maintaining control over usability, tone, and visual hierarchy.
AI functioned as a creative accelerator rather than a decision-maker, helping surface possibilities that were then shaped through intentional UX and interaction choices.
