Case study
UX Design · Product Design · Desktop
A desktop workspace for product teams that turns meeting recordings into a shared transcript, captured decisions, and assigned tasks. View live prototype
Product Designer & Developer · Prototype · 2026 · Solo designer-developer · Desktop web
Tools·Figma · Cursor · Next.js
Opal started from a gap I kept hitting as a software engineer: there was no clean way to summarize a meeting from everyone’s point of view, or to turn a decision into a task without a tedious multi-step process. Notes went into docs, recaps went out over email, and tickets were created somewhere else — so the record of the meeting was scattered. I designed and built a single desktop workspace that records and transcribes meetings with AI, attaches tasks to those decisions, and lets engineering, design, and product share one dashboard instead of three follow-up rituals.
Notes, email, and tasks splinter. People capture the meeting in a doc, send a recap, then recreate the work in a tracker. Nothing is the source of truth.
No shared summary. There isn’t a streamlined way to see the discussion from all contributors’ perspectives — engineering, design, and product each leave with a different version.
Decisions don’t become work. Turning what was agreed into an assigned task takes too many steps, so the administrative burden sits on whoever cares enough to do it by hand.
Cut that overhead with one tool: AI transcription of recordings, tasks created from decisions, a central dashboard to manage them, and a way to assign work to projects and the people across functions who own it.
I framed both the user problem (scattered notes, unpaid admin work) and the business problem (teams stay misaligned because follow-through is expensive). The design had to serve cross-functional contributors — designers, PMs, and developers — not a single role’s note-taking habit. I used my own engineering workflow as the brief, then looked at how existing boards and meeting tools split the same loop apart.
Wrote the user and business jobs from firsthand experience: reduce administrative overhead, and make eng / design / product look at the same record after a call.
Sketched the path from recording → transcript → decision → task. Modeled the task surface after Jira-style boards, because that’s where the work already tried to live — just not connected to the meeting.
Listed what a product contributor actually needs to stay productive: daily meetings, a home dashboard, projects, people, files, and more than one way to see tasks. The hard part was not adding everything — it was naming the set.
places work used to scatter
task views for different working styles
live prototype built in Cursor
32 · Product Manager · San Francisco
“I leave a meeting sure we decided something, and two days later the notes are in a doc, the recap is in email, and the tickets still aren’t filed.”
The process was linear on purpose: frame → lo-fi in Figma → moodboard → hi-fi → working prototype in Cursor, then iterate on what the first build was missing. I collaborated with ChatGPT on visual direction while I locked structure myself, so AI helped explore look, not invent the IA.
The rail is the product: Home, Meetings, Tasks, Projects, People, Settings. Home is the central dashboard. Meetings is the daily breakdown — up next, calendar, today’s table. Tasks needed board, list, and calendar so different contributors can use the same work. Global search sits above all of it, because finding a meeting, task, project, or file should not depend on remembering which tab it lives in. People exists so engineering, design, and product are named in the system, not only in the recap email.
I sketched low-fidelity wireframes in Figma before any color: a dark sidebar for the jobs, a three-up meetings canvas (next event, calendar, today’s counts), and a table of the day underneath. That layout was the contract for “scan today, join what’s next, open notes.” Decisions sat on the meeting itself — owner, status, create task — so a call could become work without leaving the room. The task board was modeled on Jira so the destination for a decision would feel familiar. Hi-fi kept this skeleton and filled it — real calendar, people, Join meeting as the one solid action.
After the lo-fi, I built a visual style moodboard and used ChatGPT to explore directions: consumer notes apps (pastel glance cards, lilac calendars) against serious work software (black rail, white settings, named people on a meeting card). Opal is the overlap — gem-colored enough to feel like a product, structured enough to sit next to a Jira-like board all day.
The moodboard’s purple became the brand mark and the calendar’s today state, not a wash over the whole UI. Black rail / white canvas came from the lo-fi and from the settings reference — it keeps the product feeling like software, not a notes toy. Glance cards stay pastel and separate (blue, mint, peach) so Home can be read in one pass, the way the phone mockup counted meetings, notes, and time.
Rail
#0E0F13
Sidebar, primary buttons
Opal
#7C5CF6
Brand, selected nav, today on the calendar
Canvas
#F4F5F8
App background
Glance mint
#E6F6E8
Open tasks at a glance
Glance peach
#FCF6F0
Needs review — work that is waiting
Titles are large and tight, like the wireframe’s Good Morning, Alex. Body copy stays small and cool-gray so a day’s table can hold names, times, and notes without shouting. Purple is reserved for selection and “up next” — never for every chip.
Inter
UI, tables, and meeting titles — dense enough for a desktop work surface
“Good morning, Alex.” “Here’s what’s on your schedule today.” “Join meeting.” The product talks like a desk, not a coach.
Hi-fidelity started as Figma mockups, then I used Cursor to build a working prototype. Meetings keeps the wireframe’s three-up scan. Home is the central dashboard the first build didn’t have. Tasks ship as board, list, and calendar, because one view would have locked the product to one kind of contributor. The live prototype is the demo — including AI-transcribed notes that can become assigned work.
One loop instead of three tools: record and transcribe, make the decision a task, manage it on a dashboard, assign it to a project and a person.
Recordings transcribe into notes on the meeting. A decision can become a task there — no doc, no recap email, no second tracker to re-type it into.
Home is the glance across meetings, tasks, and projects. Meetings is today: up next, calendar, and a table of the rest. Join and notes stay attached to the event.
Different contributors organize work differently. The task surface ships three views so a designer, a PM, and a developer can all use the same board.
Work is assigned to projects and to named people across functions. Files sit on the meeting. Global search finds a meeting, task, project, or file without hunting tabs.
The first Cursor prototype was useful and incomplete. I treated the gaps as a second design pass, not as polish.
working prototype from Figma into Cursor
task views for different contributors
design passes after the first build
Two challenges sat on top of the product ones. Guiding Cursor meant rewriting prompts until they were modular and explicit — interactions, hover states, responsive layout, visual consistency — or the generation drifted. And I had to map the exact feature set a product contributor needs to stay productive, without turning the rail into a junk drawer. If I had more time I would thicken the design system, go deeper on AI-assisted project management (permissions, org-wide workspaces, an analytics home for completion trends), and cut sub-menus so the hierarchy stays concise.
Building the working prototype — Figma to Cursor — so the demo is the product: record, transcribe, task, assign, search, instead of a slide of the idea.
Next