Hami Chat

At BSI, I partnered with data scientists and engineers to design Standard Discovery — an AI-powered chat tool that helps users quickly find and interpret standards.

My role went beyond UI/UX design. I:
Designed the chat experience end-to-end, from wireframes to high-fidelity prototypes.
Facilitated workshops with data scientists and engineers to shape implementation.
Introduced business strategies, such as subscription-based upsell mechanisms.
Delivered a full prototype for developer handoff, ensuring smooth translation into build.

Team
2 Data Scientist
2 Engineers
Project manager

My Role
Product Designer

Duration
3 Months

Screenshot of a chat interface titled 'Hami Chat' with the message 'Hello Jessica! How can I help you today?' and options for starting a new chat, customization, or browsing the library, set against a light gray background.

The Challenge

Standards are complex and technical, making them hard for users to navigate. Searching through documents often took hours, and interpreting clauses without guidance was even harder.

Meanwhile, tools like ChatGPT have raised expectations — users now demand instant, conversational answers. But with BSI’s vast library of standards, keyword overlaps made accuracy a real challenge, raising the risk of hallucinations or irrelevant results.

For the business, there was a parallel challenge: building a solution that didn’t just improve user experience but also created economic value and fit the roadmap for AI adoption.

Comparison of two chat interface screens. Left screen features a light background with a conversation about ISO 9001 standards, including questions, a reliability score bar, and informational text. Right screen shows a dark-themed interface with folder navigation, chat list, and a conversation containing placeholder text and two citation links.
A group of five people in a meeting room, four are seated and one is standing near a whiteboard, engaged in discussion.

Users

We identified three core personas:

  • Small business owners / SME managers
    Limited resources, need quick, affordable compliance help.

  • Corporate compliance managers
    Coordinating global compliance and regulatory audits.

  • Educators & researchers
     Using standards to teach and underpin research

    Across all groups, users wanted speed, clarity, and trust.

My Approach

To address both user needs and business goals, I worked in three parallel tracks:

Design Leadership

Created the chat interface and workflows, iterating from sketches → wireframes → high-fidelity prototypes.

Designed the dashboard integration, foldering system, and scoped search filters.

A collage of screenshots of website pages, including terms of use, a survey or questionnaire, safety management, and user feedback forms.

Collaboration & Workshops

Ran workshops with data scientists and engineers to align on how to reduce hallucinations (citations, reliability scores, prompt engineering).

Worked with stakeholders to ensure features aligned with the long-term roadmap.

Business Strategy

Proposed the subscription constraint mechanism: if a user asked about a standard outside their plan, the AI would acknowledge the answer exists but lock access until they upgraded. This created transparency for users and a built-in upsell pathway for the business.

A computer screen displaying a chat message about forecast reliability scores, with a gray background including a blurred flowchart in the upper left corner.

Design & Features I Delivered

  • Conversational Chat: ChatGPT-style interaction, but tailored for standards.

  • Reliability Indicators: Every answer included citations and a reliability score.

  • Personalization: Onboarding questions (industry, company size, role) for tailored responses.

  • Scoped Search: Users could filter results by subscription library or individual documents.

  • Foldering System: Saved chats for future reference and organization.

  • Upsell Mechanism: Subscription constraints turned locked answers into natural upgrade prompts.

A screenshot of an online chat or survey interface with a conversation about understanding needs, displaying a question and multiple-choice options.
Screenshot of an online assessment interface showing a question about the size of an organization with options: -30, 100-400, +400, not sure. There is a profile illustration and a progress bar indicating 85% completion, with options to save or exit.
A web browser window displaying a glossary of terms related to metals and hydrogen with a sidebar on the left and a main content area on the right.
A computer screen displaying a user interface with options for managing safety standards and selections, including folders labeled 'Safety Management' and an eye graphic on the right side.

Forecasted Impact

Forecasted Impact (Based on Industry Benchmarks)

Efficiency: With AI handling ~50–80% of routine queries, we expect to reduce average search time by 40–60%, freeing up user time and increasing satisfaction.

Engagement: Given high chatbot adoption trends, we anticipate over 60% of logged-in users will try Standard Discovery post-launch.

Business Value: AI-driven upsell moments may boost subscription expansion by an estimated 10–20%, reflecting how chatbot insights can directly influence revenue.

Trust & Retention: Including reliability scoring and citations builds trust, aligning with broader trends showing increased retention and loyalty for AI-supported experiences.

Reflectiont

This project reinforced that trust is the foundation of AI UX. By designing for transparency (citations, reliability scores) and control (scoped search, personalization), users felt empowered rather than misled.

It also showed me how UX can drive business growth. My subscription constraint idea turned a design challenge into a revenue opportunity, strengthening the link between product usability and commercial impact.

Finally, facilitating workshops with data scientists taught me the value of cross-disciplinary collaboration — aligning design vision with technical feasibility early on.

Next steps will focus on user testing, refining the reliability scoring, and expanding personalization features.