Mobile App
Ghostmeet
Teams were drowning in meetings with no reliable way to capture what was decided or who owns what. I designed Ghostmeet to solve this at the workflow level — an AI assistant that joins meetings autonomously, transcribes in real time, and surfaces structured summaries. The core design challenge was making autonomous AI actions feel transparent and trustworthy without requiring users to babysit the system.
Design Systems
Rapid Prototyping
AI & Data Visualization
User Research
User Research
Responsive Web
Responsive Web
ROLE
Ghostmeet is my own startup. I served as the founding designer and sole product owner — responsible for every design decision from initial concept to shipped product. I worked directly with engineering to navigate technical constraints and iterated rapidly based on real user feedback.
RESULTS
Shipped 15+ product iterations in 3 months, growing to $2k MRR with paying customers across web, CLI, and Slack. Reduced meeting overhead for users by automating transcription, summarization, and action item capture through an AI-native, multi-surface experience.
Project overview
Teams were drowning in meetings with no reliable way to capture what was decided or who owns what. I designed Ghostmeet to solve this at the workflow level — an AI assistant that joins meetings autonomously, transcribes in real time, and surfaces structured summaries. The core design challenge was making autonomous AI actions feel transparent and trustworthy without requiring users to babysit the system.
Design Challenge
1
Defining the Problem in Real Time
The idea came from my own frustration — I was losing track of decisions and action items across back-to-back meetings. I started with 8 user interviews across PMs, engineers, and founders to validate whether this was systemic. Every single person described the same core pain: meetings happened, things were decided, and then nothing was captured reliably. That insight shaped the entire product direction.
2
Choosing the Right Platform & Stack
Early on I had to make a real architectural decision: where does an AI meeting assistant actually live? Browser extension, desktop app, mobile, or CLI? Through user interviews it became clear that professionals wanted something invisible — it had to join meetings autonomously without adding another tab or tool to manage. That led to the decision to build a bot-based architecture that integrates at the calendar and conferencing layer, with Slack as the primary output surface.
3
Designing the Information Architecture
The hardest UX challenge was structuring the output. Raw transcripts are useless — users needed summaries, but summaries lose nuance. I designed a three-layer output system: a short TL;DR, structured action items with owners, and a searchable full transcript beneath. This hierarchy meant users could get value in 10 seconds or go deep when needed.
4
Iterating Toward Trust
The biggest friction wasn't functionality — it was trust. Users were uncomfortable with an AI bot silently in their meetings. I ran rapid prototype iterations testing different transparency patterns: pre-meeting notifications, in-meeting status indicators, and post-meeting confirmation flows. The final system gives users clear signals at every stage so the AI feels like a teammate, not surveillance.
The Impact of a Cohesive Design System
Building a unified design system taught me how consistency boosts both user efficiency and design team productivity.
Balancing Usability with Compliance
Designing for healthcare showed me how to simplify complex information while meeting strict industry standards.
Turning Data into Usable Insights
Collaborating with Data Science helped me make complex data actionable through AI-driven visualizations.
Value of Real-World Testing
Iterating based on feedback reinforced the importance of adapting designs to fit users’ real workflows.
Power of Cross-Functional Collaboration
Working with engineers and researchers highlighted how teamwork strengthens design outcomes from concept to launch.
Terms of use
Privacy police