{"id":126,"date":"2024-08-06T01:23:03","date_gmt":"2024-08-06T04:23:03","guid":{"rendered":"https:\/\/brunopavani.com\/?page_id=126"},"modified":"2026-04-27T07:39:38","modified_gmt":"2026-04-27T10:39:38","slug":"how-to-develop-an-e-commerce-for-multiple-innovative-products","status":"publish","type":"page","link":"https:\/\/brunopavani.com\/?page_id=126","title":{"rendered":"Lumia IA \u2014 Clinical AI Assistant"},"content":{"rendered":"\n

Designing a human-centered clinical decision support experience grounded in physician research and AI safety requirements<\/p>\n\n\n\n

Project Snapshot<\/h3>\n\n\n\n

Product:<\/strong> Lumia IA \u2014 clinical decision support assistant (concept-to-design)<\/p>\n\n\n\n

Timeline:<\/strong> Dec 2025 \u2013 Jan 2026<\/p>\n\n\n\n

Role:<\/strong> Lead UX\/UI Designer (end-to-end)<\/p>\n\n\n\n

Stakeholders:<\/strong> AI specialists (requesters) + practicing physicians (interviews\/feedback)<\/p>\n\n\n\n

Scope (What I designed):<\/strong> trust-first UI (sources\/confidence\/limitations), review-and-confirm workflow, auditability, safe language, reusable patterns for scalability<\/p>\n\n\n\n

Outcomes (validation, illustrative):<\/strong> ~25\u201330% faster time-to-orientation; +15\u201320% perceived trust uplift<\/p>\n\n\n\n

Confidentiality:<\/strong> Concept + early validation; metrics illustrative.<\/p>\n\n\n\n

\n\n\n\n

Summary<\/h3>\n\n\n\n

Lumia IA is a concept for a clinical AI assistant designed to support physicians during diagnostic decision-making.<\/p>\n\n\n\n

In clinical environments, time is limited and decisions carry real consequences. Physicians don\u2019t need more information \u2014 they need clarity, context, and confidence.<\/p>\n\n\n\n

The challenge was to design an AI-assisted experience that speeds up understanding without removing control, keeping the decision clearly human while making reasoning transparent.<\/p>\n\n\n\n

I led the UX\/UI design end-to-end, working with physicians and AI specialists to shape an experience where AI supports thinking, rather than replacing it.<\/p>\n\n\n\n

Context<\/h3>\n\n\n\n

Physicians often work under time pressure, navigating incomplete information and multiple possible diagnoses.<\/p>\n\n\n\n

In this context, an AI assistant can either reduce cognitive load \u2014 or increase it.<\/p>\n\n\n\n

If suggestions are unclear, overly confident, or disconnected from evidence, trust drops immediately.<\/p>\n\n\n\n

Designing for this environment means respecting how doctors think: quickly scanning, focusing on signals, validating assumptions, and documenting decisions.<\/p>\n\n\n\n

My Role<\/h3>\n\n\n\n

Lead UX\/UI Designer working with the AI specialists who requested the project and interviewing practicing physicians to ground the product in real workflow needs.<\/p>\n\n\n\n

The Design Challenge<\/h3>\n\n\n\n

The core challenge was not presenting AI suggestions \u2014 but making them usable in a clinical context.<\/p>\n\n\n\n

The experience needed to:<\/p>\n\n\n\n

    \n
  • support fast understanding without oversimplifying<\/li>\n\n\n\n
  • make reasoning visible without overwhelming<\/li>\n\n\n\n
  • communicate uncertainty without reducing trust<\/li>\n\n\n\n
  • keep the physician fully in control<\/li>\n<\/ul>\n\n\n\n

    Constraints and approach<\/h3>\n\n\n\n

    Instead of designing around \u201cAI output,\u201d I designed around the clinician\u2019s mental flow:<\/p>\n\n\n\n

    skim \u2192 focus \u2192 verify \u2192 decide<\/strong><\/p>\n\n\n\n

    The interface was structured to follow this sequence, ensuring that information appears at the right moment, in the right level of detail.<\/p>\n\n\n\n

    At the same time, AI constraints such as explainability, traceability, and safe communication were embedded directly into the interaction model.<\/p>\n\n\n\n

    Discovery and research<\/h3>\n\n\n\n

    I grounded the design in two perspectives:<\/p>\n\n\n\n

    Physician interviews<\/strong><\/p>\n\n\n\n

      \n
    • Interviewed physicians to understand how they confirm diagnoses, where they lose time, and what would make an assistant genuinely usable.<\/li>\n\n\n\n
    • Key themes (illustrative but realistic):\n
        \n
      • Fast access to the \u201cwhy\u201d matters more than the final answer<\/li>\n\n\n\n
      • Clear separation between patient facts<\/em> vs AI interpretation<\/em> builds trust<\/li>\n\n\n\n
      • Doctors want assistance with summarization, differential hypotheses, and red flags\u2014while keeping authorship of the decision<\/li>\n<\/ul>\n<\/li>\n<\/ul>\n\n\n\n

        AI specialist interviews (project requesters)<\/strong><\/p>\n\n\n\n

          \n
        • Interviewed AI specialists to define constraints for safety, explainability, and responsible use.<\/li>\n\n\n\n
        • Key themes:\n
            \n
          • Traceability is required for internal review and governance<\/li>\n\n\n\n
          • Confidence must be communicated carefully (avoid false certainty)<\/li>\n\n\n\n
          • The interface needs structured feedback loops to improve model behavior safely<\/li>\n<\/ul>\n<\/li>\n<\/ul>\n\n\n\n

            Key improvements (iteration highlights)<\/h3>\n\n\n\n

            1. Clinician-first information hierarchy<\/strong><\/h3>\n\n\n\n

            The interface prioritizes what matters in sequence:<\/p>\n\n\n\n

              \n
            • Patient overview with key signals<\/li>\n\n\n\n
            • Clinical flags and missing critical data<\/li>\n\n\n\n
            • Diagnostic hypotheses (not conclusions)<\/li>\n\n\n\n
            • Evidence and reasoning linked to each suggestion<\/li>\n<\/ul>\n\n\n\n

              The goal was to reduce the time to \u201cunderstand the case\u201d without hiding complexity.<\/p>\n\n\n\n

              \n\n\n\n

              2. Suggestions designed as hypotheses<\/strong><\/h3>\n\n\n\n

              Instead of presenting answers, the system presents possibilities<\/strong>.<\/p>\n\n\n\n

                \n
              • Confidence shown as ranges, not certainty<\/li>\n\n\n\n
              • Clear \u201cwhy\u201d behind each suggestion<\/li>\n\n\n\n
              • Evidence always accessible at decision moment<\/li>\n<\/ul>\n\n\n\n

                This avoids overtrust while supporting faster reasoning.<\/p>\n\n\n\n

                \n\n\n\n

                3. Review-and-confirm interaction model<\/strong><\/h3>\n\n\n\n

                The workflow mirrors real clinical thinking:<\/p>\n\n\n\n

                  \n
                • Compare hypotheses<\/li>\n\n\n\n
                • Validate evidence<\/li>\n\n\n\n
                • Confirm a direction<\/li>\n\n\n\n
                • Document reasoning<\/li>\n<\/ul>\n\n\n\n

                  The system supports decisions \u2014 it doesn\u2019t make them.<\/p>\n\n\n\n

                  \n\n\n\n

                  4. Feedback loop without friction<\/strong><\/h3>\n\n\n\n

                  Feedback is embedded into the flow:<\/p>\n\n\n\n

                    \n
                  • Quick \u201chelpful \/ not helpful\u201d<\/li>\n\n\n\n
                  • Option to flag issues<\/li>\n\n\n\n
                  • Lightweight structured input<\/li>\n<\/ul>\n\n\n\n

                    Improving the model without slowing the clinician down.<\/p>\n\n\n\n

                    \n\n\n\n

                    5. Visual language: calm, clinical, human<\/strong><\/h3>\n\n\n\n

                    The UI avoids \u201cAI aesthetics\u201d and focuses on trust:<\/p>\n\n\n\n

                      \n
                    • Clean hierarchy and restrained colors<\/li>\n\n\n\n
                    • Soft visual elements to reduce tension<\/li>\n\n\n\n
                    • No \u201cmagic\u201d or exaggerated AI signals<\/li>\n<\/ul>\n\n\n\n

                      The system feels reliable, not experimental.<\/p>\n\n\n\n

                      Approved layout<\/h3>\n\n\n\n

                      Final layouts designed by me, reviewed with AI specialists and validated through physician feedback sessions.<\/p>\n\n\n\n