The birth of Pi: Bringing emotional intelligence to AI
2023 • iOS & Web
Pi set out to redefine AI agents with an emotionally intelligent assistant. During it's quick venture from MVP to fully functional app, we worked on core chat features, the launch of voice calls, and a scalable design system. We built at speed, testing in real-time, which ultimately helped shape a product that was intuitive and natural, that was swiftly acquired by Microsoft.
Intro
In 2022, Mustafa Suleyman, co-founder of DeepMind, and Reid Hoffman, co-founder of LinkedIn, launched Inflection AI, a company dedicated to reshaping how humans interact with AI. Their flagship product, Pi(short for personal intelligence), was built to engage in natural, empathetic conversations, setting it apart from other AI assistants.
Unlike competitors such as ChatGPT, Claude, Perplexity, Bing, and Co-Pilot, Pi wasn’t just about delivering information; it was about emotional intelligence, guiding users through thoughts, ideas, and personal reflections in a way that felt supportive and conversational rather than transactional.
AI was moving fast, and so was Inflection. Our goal was simple—to create a seamless, intuitive experience that encouraged open exploration. One that helps users express, reflect, and engage with AI in a way that felt meaningful.
This was a zero-to-one product. Taking an entirely new conversational AI platform to market at that pace, while shaping user-facing experiences from the ground up, was a huge undertaking—and a highlight of my career. I joined the team halfway through the initial launch phase and had a direct hand in shaping the product from its earliest form—defining beyond MVP and layering on meaningful functionality through iterative releases.
What I worked on & My impact
I was one of the first designers to join the team, alongside another designer, a design lead, and a small iOS team(their internal team were handling web and expanded to android later down the line).
I joined just as the first MVP shipped, and helped define and deliver the next iteration—building on core functionality and layering in new features to increase value and drive engagement.
Pi was built on a series of progressive LLMs, designed to support users in different mental models; but right from the start, empathy drove everything, and that is a design philosophy which I could easily get behind.
Alongside product definition, I also led the development of Pi’s scaled design system—building token-based architecture ensuring more consistent styling and to ensure the new brand rollout landed effectively and up to standards. This foundational work enabled us to ship even more quickly as the product evolved.
We focused on shaping its IA, defining key use cases, and surfacing new functionality to enhance both user experience and model training. When I joined the team the product was a simple chat interface, but we gradually expanded its capabilities, continuously testing patterns and refining ideas at pace.
Key features
Call Pi – Allowed users to speak with Pi via voice, enabling on-the-go conversations. This was a unique offering until ChatGPT introduced a similar functionality.
Mode Selection – Enabled users to communicate with Pi in different styles and tones, helping to both train the AI and improve user experience.
Discover – A quick-start feature with curated prompts, helping users engage with Pi in new ways.
Share – Allowed users to share responses with new or existing users.
Rate or Report Responses – A feedback mechanism that improved model accuracy and safety.
Working with a San Francisco startup was fast-paced and exciting. We were shipping and learning weekly, iterating quickly based on real user behaviour. Developing a new AI model in real time meant reacting quickly to solve problem as and when they emerged, but with a rapidly growing user base, we could reach statistical significance on A/B (and sometimes C) tests at speed.
We had regular check-ins with Inflection's heads of design, sometimes daily, to ensure design excellence was upheld throughout the experience. It was invigorating to collaborate with people deeply committed to building something truly best-in-class and bringing a wealth of design expertise. Although they had not actually designed a digital product before, as they were brand focused, the combination of us as product experts and them as visual visionaries allowed us to create something quite magical, whilst sharing valuable knowledge across the pond.
Scaling Pi’s design system
Alongside product design, I also led the design system and style library, ensuring Pi was built to scale. A significant part of this was:
Integrating dark mode for iOS and Android
Building a design-mapped token component library
Implementing the evolving brand across our work
Working closely with developers, we ensured consistency across platforms, creating a scalable system that supported Pi’s rapid growth.
Lessons in rapid AI development
This project moved incredibly fast, and I learned that you don’t always need to overthink decisions when you have a strong testing, iteration loops and regular client feedback. Because we had the ability to roll back quickly and test carefully, we could test in live environments without fear of causing disruption.
DESIGN tokens mapped out for swift brand updates
the COLOUR tOKEN SYSTEM MAPPED WITHIN FIGMA
Impact & Key data points
