Maike

AI-Powered Citizen Development Platform

March 2024 - June 2024

Role
Lead Product Designer

Client
BP (Incubation)

Scope
Could AI let staff build their own tools, saving the cost of a product team?

Outcome
Employees prototyping and validating their own ideas, no product teams required

Summary

From prototype to a community that wouldn't let go

Maike was an AI platform that enabled non-technical BP employees to create their own tools without needing a development team.

When I joined, it was functional but felt more like a developer experiment. My goal was to make it feel trustworthy and accessible. I developed a compact design system, worked on generative UI that transformed raw AI outputs into familiar formats and components, and contributed to shaping a platform that was LLM-agnostic and multimodal, supporting text, documents, images, audio, and video.

The primary focus was always on making powerful tools approachable. We built a community around Maike through weekly calls, onboarding sessions, and a Teams channel with over 100 members. Even after the incubation team was stood down, BP continued to operate Maike, with the community actively contributing to its development.

Screenshot of a software platform's dashboard with a dark theme, featuring a top navigation bar, a large banner titled 'Getting started' with an illustrated background of a person looking at a landscape with trees and glowing icons, and sections labeled 'Featured' and 'More tools' with various tool options and descriptions.

The Challenge

Maike had generated interest but felt like a developer experiment. Without stronger foundations, it risked being seen as a technical curiosity rather than a credible tool. My role was to give it a cohesive design identity and make it approachable for people who'd never used developer tools.

Screenshot of a web interface for an AI tool titled 'My new AI Tool' with input fields for name and prompt, listing various global issues such as rural internet access, financial services, mental health, healthcare, education, childcare, clean energy, online learning, and eco-friendly products.

The product, when I joined, was functional but unpolished. No design system, no visual identity.

Making AI Approachable

Generative AI was gaining popularity, but it felt clunky at the time. Maike was trying to go beyond a conversational interface. I designed a set of components that could format raw AI outputs into familiar forms, fields, and previews. User testing confirmed that showing what was happening behind the scenes made people trust the results more than instant answers.

Screenshot of a software interface titled 'New Maike' offering tools such as AI persona generator, multi-species image analyzer, advanced URL extractor from text, and user interview thematic extractor. The central question is 'What tool would you like to make today?', with a text box featuring the user's request for an NDA scan for clauses that could put BP at risk, and a 'Generate tool' button beside it.

Prompt suggestions helped users get started.

Progress screen displaying steps in a process: Processing requirements, understanding needs, connecting to UI library, selecting UX pattern, running usability test, optimizing design, sending algorithms.

Showing the build process helped them trust the results.

Flexibility by Design

The platform was designed to be LLM-agnostic and multimodal. Users could switch between models depending on their use case, and bring in text, documents, images, audio, and video; each input type felt consistent and approachable.

Screenshot of a computer screen showing advanced settings for selecting a GPT model, with options including Azure GPT-4, GPT-4-1106 preview, GPT-4-32k, Llama-3-8b-instruct, Phi-3-mini-4k-instruct, and Mistral-7b-instruct-v0.3.

Users could switch between models depending on their use case — Azure GPT-4, Llama, Mistral, and more.

Generative UI

Screenshot of an NDA scanning software interface showing a completed scan, NDA details, insights including scanned clauses and issues detected, and options for managing key clauses and potential risks.

A complete tool generated from a single prompt — upload, scan, review results.

Towards the end of my time on Maike, we introduced generative UI, which the platform could create complete interfaces from a single prompt. Users could publish tools company-wide in editable or run-only modes, leading to truly no-code tool creation.

Building a Community

Collection of digital tool icons and descriptions for tools like Automated NDA Scanner, UX Research Synthesis Tool, Product Spec Standardizer, Private Document Q&A, Strategic Plan Analyzer, Discovery Coach Bot, Discovery Learning Quiz, Ticket Pattern Analyzer, and Journey Synthesis Tool. Organized in a grid with labels like 'Featured' and 'More tools'.

Tools built by the community spanned legal, UX research, data processing, and strategy.

We didn't just release Maike, we built a feedback loop. A Teams channel grew to 100+ members, with weekly calls drawing around 50 participants. A dozen power users built tools daily. When the incubation team was stood down, the community kept going.

Outcomes

100+

community members

12

Power users building daily

50

Weekly call participants

12+

Months active after the team stood down

Reflection: Adoption comes from clarity, not hype. By grounding the tool in approachable interfaces and building an engaged community, we showed how design can turn cutting-edge AI from a concept into practical, day-to-day value.