01 · Challenge
Powerful docs, slow answers
Number's payment platform comes with an extensive documentation library. Finding the right page in it took real time, and that friction landed in the middle of every integration. We proposed and built a way out: a conversational assistant living inside the docs themselves.
No room for wrong answers
For a payments provider, incorrect integration advice is not an option. Hallucinated answers would be worse than no assistant at all. Every response had to be reliable, verifiable, and sourced directly from the official documentation.
Answering is not accelerating
A plain Q&A bot wouldn't move integration timelines. Developers needed ready-to-use code and a clear path into the deeper material. And the pre-built Azure RAG tool that fit the rapid timeline came with a generic UI that clashed with Number's brand.
02 · Solution
3 layers between question and answer
Prompts that stay inside the docs
A rigorous prompt engineering strategy constrains the AI's responses strictly to the source documentation. Every answer includes linked citations, so developers can click through to the source page, verify the information, and build trust in the tool.
A front end that looks like Number
Our designer and developer worked in tandem to apply a custom front-end layer over the Azure tool. Number's design system—typography, colors, interface elements—carries through, so the assistant feels like a native part of the developer portal rather than a bolted-on widget.
Code you can paste, links that keep your place
The assistant recognizes when a code example is the real answer and pastes formatted snippets directly into the chat. Source links open in new tabs, so developers get what they need and dive deeper without losing their place mid-integration.
03 · Results
Onboarding measured in minutes
Faster developer onboarding
Developers get accurate, context-specific integration details on demand instead of searching large volumes of documentation by hand. New team members now understand the available payment methods for each integration type in minutes rather than hours or days.
Consistent payment-method mapping
Before the assistant, developers cross-referenced multiple sources to match payment methods to integration channels—slow, and prone to inconsistent implementations between teams. Source-linked answers cut integration errors, rework, and the compliance risk that came with misreads of security and data-handling requirements.