500 Hours of Audio. 120 Artists. Zero Spreadsheets.
How we automated artist onboarding and project delivery time by 35%.
Role
UX Designer
Industry
B2B, SaaS, Workflow
Customer
Murf AI, Deepgram, AiOla & other Frontier AI Labs
This Project Comes under NDA, for that it covers the surface of the project.
Context
Databrewery is a data labelling and production platform designed to manage, label, and deliver AI-ready datasets. Databrewery's ops team was managing audio production for frontier AI clients, projects requiring hundreds of hours of voice recordings across dozens of languages and accents. At any given time, they were coordinating with a global network of 700+ artists, tracking assignments in Excel, communicating via WhatsApp and email, and manually syncing everything back into the platform. It worked, until it didn't. When a 500-hour project came in with a 14-day deadline, the cracks became craters. The ops lead told us: "We're spending more time updating spreadsheets than actually managing artists." That conversation started this project.
Problem Statement
The core problem was deceptively simple, every piece of information lived in two places. Client requirements went into Excel. Artist assignments went into the platform. But when something changed and it always changed, both had to be updated manually. Two sources of truth meant one was always wrong. For artists, the experience was worse. Onboarding meant fielding messages across WhatsApp, email, and the platform, filling in personal details through separate channels, and waiting for confirmation that never felt official. New artists dropped off before completing their first assignment. Design challenge, consolidate a fragmented, multi-channel workflow into a single platform-native experience without disrupting the ops team's existing rhythm.
The Real Problem Wasn't Distribution. It Was Trust.
Frontier AI labs don't just buy data, they stake their model quality on it. The old email process gave them no way to evaluate what they were getting before committing. No preview. No metadata. Just a spreadsheet and a promise.
The design challenge was precise: build a platform that makes 40+ proprietary datasets discoverable, evaluable, and requestable, without a login wall, while creating enough transparency to earn trust from the most technical buyers in the industry.
My Role & Contribution
I was the sole designer on this project, working directly with the ops team and 4 engineers. There was no product manager, which meant I was also responsible for scoping what we built, in what order, and what we deliberately left out.
My work spanned the full design process: 3 days of contextual observation embedded with the ops team, mapping the full artist onboarding journey from client request to sample submission, identifying where the platform could absorb manual steps, and designing the replacement workflow end-to-end.
I also introduced Figma Make into the process, using AI-assisted prototyping to compress the design cycle and create space for two full rounds of testing with real artists before a single line of code was written. That decision directly contributed to catching the mobile upload gap before launch, not after.
Research & Insights
We conducted in-depth interviews with the ops team and shadowed the team for 3 days, observing how they assigned tasks and communicated with artists.
3 key insights from 3 days of shadowing:
Double entry was universal: every task existed in both Excel and the platform. There was no single source of truth.
Artist dropout happened at onboarding: new artists were asked for details across 3 different channels. Many never completed the process.
The ops team's real job was coordination, not data entry, but 60% of their time was spent on the latter.
Design & AI Prototyping
The decision to go mobile-first wasn't a trend it was a workflow reality. Most artists in Databrewery's network record audio on their phones. When studio-quality production isn't required accent samples, conversational speech, short-form recordings artists use what they have: their mobile device, a quiet room, and the built-in mic. Designing a desktop-first onboarding flow would have immediately excluded the majority of the people we were designing for.
With that constraint locked, I used Claude and Figma Make to generate the initial prototype before opening a single Figma canvas. I prompted the core flow artist registration, task briefing, sample submission and iterated in Figma Make, testing the hierarchy and step logic without committing to pixel-perfect UI. Only once the flow felt right did I bring it into Figma for refinement against the existing design system.
This compressed what would typically be a week of wireframing into two days leaving more time for what mattered most: testing with real artists.
Usability Testing & Iteration
Round 1: 3 Veteran artists, 80% aligned
We tested the prototype with a first cohort of veteran artists people already familiar with Databrewery's process, whose instincts would surface the sharpest friction. The flow worked well, registration was clear, task briefing was understood, and the submission path felt intuitive on mobile.
One thing broke immediately. Artists flagged that finding and uploading audio files on mobile is genuinely difficult, navigating file systems, locating the right recording, attaching it correctly. The friction wasn't a bug. It was the nature of mobile file management. But there was a deeper insight underneath it, because artists were recording samples specifically to meet the brief requirements, the right accent, the right tone, the right length, they weren't pulling from an existing library. They were recording fresh, right then, for this task. The most natural action wasn't upload. It was record. The fix wasn't a better file picker. We added in-platform recording, so artists could record directly inside the task, submit in one tap, and never touch their phone's file system at all.
Round 2: New cohort of 4 Artist, 99% aligned
We ran the second round with a fresh cohort artists who had never used the old workflow and had no habits to unlearn. With the recording feature in place, the flow ran end-to-end without intervention. 99% alignment across the cohort. The 1% was a labelling clarity issue on the task briefing screen, noted for V2.
What this round confirmed: the mobile-first bet was right. Every artist in the cohort completed the flow on their phone. Not one asked for a desktop link.
Final Implementation & Code Handoff
We didn't hand off a spec. We handed off working components.
Because the prototype was built in Figma Make and refined inside the existing design system, developers received production-ready front-end components alongside the final screens, no redline files, no annotation decks, no spacing debates. The handoff that typically takes a week took two days.
One constraint shaped the entire implementation: we were building inside a live platform that couldn't go offline. No full rebuild, no migration window. Every new screen had to slot into the existing component architecture without breaking what was already running. That discipline, designing for integration, not replacement, kept the rollout clean and the ops team operational throughout.
Results & Outcome
30%
Ops time reduction
120+
Artists onboarded
500 hrs
Audio Delivered AI lab onboarded
The first signal came quickly, within a month of rollout, the ops team's manual workload dropped by 35%. Tasks that previously required cross-referencing Excel and the platform were now handled entirely in one place. The double-entry problem was gone.
Then the real test arrived. A 500-hour audio project came in with a 14-day deadline, the kind of project that had previously taken a month under the old workflow. We onboarded 120 artists through the new platform, coordinated assignments, and collected all audio in 14 days. On time. At scale. Without a single Excel sheet.
That project proved something more important than the feature working: it proved the system could scale. 120 artists onboarded through a platform flow that, three months earlier, had been a chain of WhatsApp messages and manual spreadsheet entries.
"I actually managed artists today instead of data."
~Operation Lead
There's More to This Story. Let's Talk.
Due to confidentiality agreements, the full design system, internal workflows, and some product screens aren't publicly shareable. But they exist and I'd be glad to walk you through the decisions behind them.
If this project resonates with the kind of work you're hiring for, let's have that conversation.