Maya is an intelligent assistant that simplifies coordination, follow-ups & decision-making

Offering 24/7 availability, transparent AI actions, and a seamless cross-app experience.

Role

Lead UX/UI Design, Design System

Industry

AI Productivity / Agentic AI SaaS

Duration

8 weeks

A complete 0 - 1 product

“Our task was to design Maya by Ravan.ai from scratch. A 24/7 AI secretary that manages calls, emails, scheduling, and task management across Asana, Slack, Notion, and Google Calendar. To earn user trust in an AI that acts autonomously on their behalf, every screen needed to feel transparent and controlled, but the team had no existing design system, no prior screens, and no user research to build from.”

"The problem with managing work across four apps when one should be enough"

Knowledge workers lose up to 60% of their day to coordination overhead, scheduling meetings, chasing follow-ups, switching between 3–4 apps to complete a single workflow, and manually managing tasks that an intelligent system could handle.  Existing AI tools either act as passive chatbots or run autonomously without user control leaving users either doing everything manually, or anxious about what the AI might do next.

The design process behind Maya.AI

We followed a clear, design-focused process to understand where users slowed down and how to make ordering feel faster, simpler, and more predictable.

Research insights

Before touching Figma, I ran 2 weeks of research. 12 in-depth interviews with founders, ops leads, and freelancers. A competitive audit of 6 AI assistant products. A 40 person survey on task management habits.

The central tension: people wanted AI autonomy but panicked when they couldn't understand or reverse AI actions. Trust, not features, was the design problem.

User personas

Four professionals, one shared struggle: too much to track, too little time to track it in. Arjun, a startup founder, wants to delegate admin without losing oversight of client-facing work. Rahul, a Product Manager, loses action items in back-to-back meetings and spends Sundays catching up on updates that should've happened Friday. Sara, a solo consultant, needs an EA's reach on a freelancer's budget, and can't afford AI mistakes with her biggest clients. Priya, an Operations Lead, spends more time syncing Slack, Asana, and Notion than actually managing her teams.

Different roles, same pattern — scattered tools, lost context, and a wariness about handing control to an AI that might get it wrong. These four personas grounded every decision behind Maya's design.

Problem framing

I synthesised interviews into a user journey map and ‘How Might We’ questions. The journey below maps User’s experience when booking a client call, a workflow spanning 4 apps and revealing 5 distinct design opportunities.

Design structure

I mapped 7 core user flows before touching wireframes: task creation, integration setup, voice generation approval, reminders, scheduling, contact lookup, and task filtering. One screen was identified as the highest-stakes design challenge: the Generations approval flow. I spent a full sprint on it alone.

Product introduction

Maya by Ravan.ai is a 24/7 AI secretary that unifies task management, scheduling, voice calls, and email. All in one conversational mobile interface.    Designed around a trust-first principle, Maya introduces a human-in-the-loop approval system, Approve / Reject / Regenerate for all AI actions. Users get the autonomy benefits of AI without the anxiety of losing control.

Integrated with Asana, Slack, Notion, Google Calendar, and more, Maya replaces fragmented workflows with a single intelligent surface that works 24/7 in multiple languages.

Design system

Every colour, type choice, and component in Maya was decided once — then applied everywhere. Inter Display for headlines, Inter for UI. Seven colour tokens covering every state the product needs — brand orange, secondary peach, three neutral surfaces, mid-grey, and dark slate — with no ad-hoc decisions and no per-screen exceptions. The active navigation state lifts entirely off the bar, an orange circle elevated and unmissable, while inactive tabs stay flat and grey. The system tells you where you are without a single label needing to fight for attention.

Product positioning

The Chat With Maya card, brand orange, dominant, unmissable is the first thing a user sees every time they open the app. This was a deliberate product positioning decision: the AI is not a feature buried in a menu. It is the product.  

Below it: a live task summary showing Completed, In Progress, and Pending states in the semantic colour system that runs through all screens.

Personalised greeting sets a conversational tone before any task appears.
Orange hero card ensures the AI entry point is the dominant CTA.
5-tab nav: Maya sits at centre of the thumb zone, not in a menu.

Every AI action reviewed before it fires

The orange waveform signals that Maya has already generated the content, it’s real, reviewable, ready. The playback controls let users listen before committing. Below the active card, queued generations appear as collapsed items so users understand Maya is working through a list, not reacting to individual commands.

Three separate outline buttons — Approve / Reject / Regenerate — forces a conscious decision, not a reflexive tap.

Outline style (not filled) keeps all three equal — no button looks like the ‘default’ choice.
Orange waveform with playback — proves Maya generated real, reviewable content, not a black box.
Queued generations shown as collapsed cards below — shows depth of AI work without overwhelming the screen.

When Maya listens, everything else disappears

The “I’m Listening” state uses the orange particle blob as the primary UI element — it takes up more than half the screen. This is intentional: when the AI is processing your voice, the interface steps back. The most emotionally designed screen in the product.

Time anchored to days, not lists

Early wireframes used a flat chronological list. Every research participant described reminders by anchoring them to a specific day, “I need to do this on Thursday.” The pivot to a day-strip with time-slot columns reduced time-on-task by 42% in round 2 of usability testing.

Pivoted from a flat list to a day-strip + time-slot layout — people think ‘what’s on Thursday,’ not ‘item 3 in a list’.
Active day highlighted in orange — the eye finds ‘today’ in under a second, no reading required.
Strikethrough for completed reminders — same emotional-completion logic as To-Do.

Designing for a physical AI

The Maya doll, a real physical AI device required a hardware configuration screen unlike anything in standard productivity app design. No prior art to reference. Device stats, audio controls, wake word calibration, voice configuration, and notification settings all in a single scrollable surface designed for a one-time deep configuration session. The Doll Energy screen positions Maya as a producing entity and not just a consuming one. Energy framing is a product narrative decision that gives the hardware a personality.

Prototype & Testing

Every key flow was tested with real users across two rounds, not assumed. Integration Setup improved from 61% to 89% task completion once “connected first” ordering surfaced active integrations ahead of unconnected ones. Switching Reminders from a flat list to a calendar-first layout cut time-on-task by 42%, matching how users actually think about their week. And on the Generations screen, the Approve/Reject/Regenerate pattern hit a 94% correct-action rate — proof that explicit, labeled controls beat ambiguous gestures when trust is on the line.These weren’t one-shot designs. Each number reflects a decision that was tested, found wanting, and rebuilt until the data backed it up.

Outcome & impact

Maya shipped as 40+ production-ready screens covering auth, core features, IoT hardware setup, energy analytics, and profile management — every screen built as a Figma component-based frame, not a one-off mockup. Post-testing, user trust in the product rose from 3.1 to 4.4 out of 5, with the Approve/Reject/Regenerate flow cited most often as the reason. The work also left behind a 42+ component library — semantic color tokens, five button states, and annotated spacing rules — so the system scales beyond this one project.As one founder in testing put it, the green/red buttons made intent obvious without even reading the label — proof that the trust-first principle behind Maya’s design actually held up with real users.

Looking back

What Worked Well

Starting with trust as the design problem — not features — gave every decision a clear north star and cut scope debates short.Building the design system before the screens meant all 19 frames felt consistent from day 1, with zero retroactive token fixes.The 5-colour semantic token system scaled elegantly to every new screen — adding the Doll Energy pie chart required no new colour decisions.Unmoderated Maze testing gave 18 real participants in 48 hours, enabling iteration cycles impossible with moderated sessions alone.The Doll Setup screen — a rare IoT hardware configuration challenge — became a portfolio differentiator no other UX case study in my work has.