Problem Statement

GSK researchers and R&D teams need a seamless way to search, analyze, and visualize complex scientific data across diverse sources. Current tools often lack integration, accessibility, and intuitive interfaces, slowing the discovery and decision-making process. The Onyx platform leverages ontology and AI as enabling technologies to unify data, provide actionable insights, and streamline workflows. This cloud-based solution empowers users to explore and consume scientific data efficiently, driving innovation and collaboration in drug development.

Conceptual Designs

Use Case 1: Ontology Creation and Expert Validation

1. Researcher integrates and analyzes diverse large-scale datasets.

2. AI generates preliminary ontology models by identifying key relationships and concepts.

3. Researcher validates and refines AI-generated models using real-world data and experimental results.

Use Case 2: AI-Driven Interaction Modeling and Experimentation

1. Researcher develops a hypothesis for a specific biological interaction or mechanism.

2. AI simulates computational experiments, generating insights to refine the hypothesis.

3. Based on promising findings, the researcher requests validation through cloud-controlled wet lab experiments.

4. AI autonomously designs the experiment, optimizes resource allocation, and manages protocols and scheduling.

Thoughts on Chat UI

Chat UI: The New Flat Design

The rise of large language models (LLMs) has made the “Chat UI” feel almost ubiquitous in generative AI applications—like the new flat design of our time. In job interviews, designers and product managers might earnestly proclaim:

“We need to deeply understand what customers need—not just what they say they want. We’re solving real problems here… and it just so happens that those problems are best solved with, you guessed it, a Chat UI.”

Some of the AI tools I genuinely admire—like Cursor—still center around a chat interface, but go further by blending more seamlessly into the user’s workflow. The result is a kind of invisible choreography where the interface steps back, and the conversation takes the lead. It’s still a chat—but now with better timing, smoother moves, and slightly less small talk.

Morphing UI & Multimodality

Interfaces won’t look the same for everyone—or even for the same person over time. As large language models move beyond text, UX will follow—blending visuals, 3D, animation, and audio into fluid, adaptive, context-aware experiences. The same idea might surface in multiple forms, either simultaneously or on demand—whether as a natural language summary, a supporting chart, or the underlying SQL—depending on what best serves the user’s goal in that moment.

For professional and prosumer users, REPL-style environments—like Jupyter or Mathematica—will continue to rise in popularity. In my work on Onyx, we’ve observed that users are naturally drawn to these kinds of interactions: they want to sketch out ideas—whether as formulas, code, or diagrams—and engage in a back-and-forth using natural language, one snippet at a time. What they don’t want is to waste time learning a new UI—whether it’s a “low-code” builder or a traditional wizard-style GUI.

New Type of UI Fatigue

Users of generative AI apps often slip into autopilot as LLMs grow more capable—copying and pasting responses without even reading them. For many use cases, this isn’t a big problem. But in high-stakes or creative contexts, it can lead to a false sense of confidence, missed nuance, or blind spots. As the UI fades into the background, the need for thoughtful engagement becomes even more critical—raising new challenges in interface design, attention management, and trust calibration.