01 The problem
Lawyers, students, and board members all face the same problem: they can participate fully in a conversation, or document it accurately, but rarely both. A missed detail in cross-examination carries different stakes than a missed point in a lecture, but the underlying tradeoff is the same — attention and documentation compete for the same time.
There was no existing product to audit or iterate on. Before the PRD existed, the founder ran a series of discovery conversations with lawyers, students, and business professionals. I designed the first version from that foundation, working directly from founder vision and PM-authored PRDs.
02 Research: user conversations before design decisions
Because I was designing from zero, there was no existing product data or feedback loop to lean on. I spoke directly with a small group of core users — mostly lawyers, board members, and students — before and during design, and produced explainer videos to gather early client reactions ahead of full builds.
"The most useful thing wouldn't be the raw transcript — it'd be a clear set of key notes and action points I could look at right after a meeting and know exactly what's required of me."
— Board of Directors member
AI summarization and key-point extraction became a core feature. The product surfaces what someone needs to act on, not just a record of what was said.
That decision sits inside a larger pipeline. The product doesn't stop at a transcript — it carries the conversation forward into something a user can act on:
03 Key decision: folders, not tags
A recording product could reasonably default to a chronological feed or a tagging system. I chose folder-based organization instead because the target users — lawyers organizing by case, students by course, board members by meeting series — needed to group recordings by an ongoing context they already think in, not by searching for one past session at a time.
04 Fixing onboarding after watching clients hit friction
The onboarding flow wasn't finalized in one pass. Watching clients move through it surfaced a specific drop-off point: permission requests appeared early, before users understood why the app needed them.
05 The core loop
Despite the range of use cases, the primary flow had to stay simple enough to use under time pressure — starting a recording in a live meeting isn't the moment for a complex UI.
Record
Start capturing with minimal friction.
Transcribe
Structured, speaker-differentiated transcript.
Organize
Filed into folders for retrieval later.
Beyond this loop, I designed the surrounding experience needed to make a transcript usable afterward:
- AI summarization, transcript management and export, drafts, audio import
- Biometric protection for sensitive recordings — relevant given legal and corporate use cases
- A dedicated cross-examination mode for structured capture of question-and-answer exchanges in legal settings
06 Process: the full flow map
The mobile app is one entry point into a larger system. Underneath the record → transcribe → organize loop sits tier-gated access, a nested organization hierarchy, an AI processing layer, and a full admin/governance layer for workspace accounts — all mapped and designed within the same four-month scope, before a single screen was drawn in detail.
07 Impact
The product was built from zero — no users, transcripts, or meetings existed before this design work.
These are current, product-wide metrics. Growth continued well past my four-month engagement through the team's ongoing work, including a campus ambassador program. I'm not claiming sole credit for this growth — what I can claim directly is that the foundational product experience the team launched and scaled on was designed by me from an empty state.
08 What this project tested
Designing from zero, across distinct user types with very different needs, meant every major decision — how sessions get organized, what onboarding asks of a first-time user, what a legal user needs that a board member doesn't — had to come from direct user conversation, not inherited precedent or existing product data. That pattern, research before decisions, ran through the whole four-month engagement.