Type / Case Study
Year / 2023–2024
Role / Lead Product Design
Context / 35K DAU · 14 Countries

TaskGPT.
The AI suite
that paid for itself.

End-to-end design of an enterprise AI productivity suite built on OpenAI, PaLM 2, and LLaMA, used by 35,000 operators daily across 14 countries. Call handling time dropped 20%. The product earned its own cost.

TaskGPT feedback dashboard: product and knowledge ratings, conversation ratings and qualitative feedback columns
Cost Reduction
20%
Daily Users
35K
Countries
14
Problem

35,000 operators. Millions of customer queries. A playbook designed before AI existed.

TaskUs agents handle complex enterprise support across 14 countries: insurance claims, fintech queries, healthcare triage. The existing tooling was built for a pre-AI world: multiple tabs, manual copy-paste, static knowledge bases. Resolution time was high. Operator cognitive load was higher.

The brief was not "add AI to the product." It was redesign the entire operator workflow assuming AI is native to every step, from query classification to response generation to quality evaluation. A full paradigm shift, not a feature addition.

"Operators should think less about finding answers and more about delivering them."
Design principle, TaskGPT v1
AI Interface Design Enterprise UX Design Systems OpenAI · PaLM 2 · LLaMA 35K DAU 14 Countries
Approach

The PCAF framework: a new language for AI-native design.

Before designing screens, I authored PCAF, the Persona-Centric Analysis Framework, to replace assumption-led research at TaskUs. Every AI interface decision traces back to an evidenced operator persona: how suggestions surface, how operators accept or override them, and how confidence is communicated without overwhelming attention.

Every screen in TaskGPT traces back to a PCAF principle. This ensured consistency across EvaluateUs, the LevelUp gamification system, and the core TaskGPT suite: three products that share an operator audience but serve very different moments in the workflow.

01 / Research

Journey mapping with 35K+ operators

Shadowed agents across 6 site locations. Mapped every tool switch, every manual lookup, every moment where the current workflow forced a wrong decision.

02 / Framework

PCAF: persona-centric analysis

Built the Persona-Centric Analysis Framework so every AI interface decision traces to an evidenced operator persona, not an assumption. Now used across all TaskUs AI products.

03 / Design

Multi-model AI suite

Designed the full product across OpenAI, PaLM 2, and LLaMA, with each model serving a different operator task with appropriate confidence indicators.

04 / Ecosystem

LevelUp gamification

Designed the LevelUp engagement layer (challenges, leaderboards, streaks) that lifted operator engagement 40% and reduced attrition on high-volume queues.

The admin plane

Operators use the AI. Admins configure it.

TaskGPT is two products in one interface. Operators live in the assistant during live customer conversations. Behind them sits the admin plane: chatbots configured per client, usage and consumption dashboards, and data-dense feedback reporting that tells the team what the AI is actually doing in production. Designing both directions at once, configuration on one side and lived workflow on the other, is what made the suite hold together.

TaskGPT admin plane: creating and configuring a chatbot for a client, with type, welcome message and JSON configuration
Client configuration · one chatbot per client, from welcome message to JSON config (client name redacted)
TaskGPT feedback report: data-dense table of ratings by operator with experience, quality and NLU scores
Feedback reporting · every rating traceable to a session, exportable for clients
TaskGPT dashboards: users, sessions, responses and token consumption overview plus feedback ratings view
Usage dashboards · users, sessions, valid responses and token consumption at a glance
Outcome

20% cost reduction.
35K daily users. One framework that changed how TaskUs builds AI products.

TaskGPT shipped across 14 countries and 35,000 daily operators. Call handling time fell 20% within the first two quarters, a reduction that translated directly to operational cost savings for enterprise clients.

The PCAF framework, originally built for TaskGPT, was adopted company-wide as the standard for AI interface decisions. EvaluateUs, the performance evaluation suite, and the LevelUp gamification layer both shipped under it. The design system outlived the project it was born from.

20%
Operational cost saved
35K
Daily active operators
40%
Engagement lift via LevelUp