Designing AI and ML for trust!

Overview

Watson for Drug Discovery was redesigned to turn search from basic retrieval into a discovery engine capable of surfacing gene–disease relationships hidden across 100+ million articles. By aligning search with how scientists form hypotheses — using guided facets, transparent evidence, and iterative exploration — we enabled breakthroughs like identifying novel ALS-related genes and cut discovery timelines from months or years to weeks.

Legacy UI showing paginated represenations that increased cognitive load.

Before: Before: The Entity Explorer experience forced users to set complex filters, remember results across many sets of prioritized entities,and guess at the names of entities from 5 letters presented to the user.

Before: Individual article views were poorly structured — with hidden labels, inaccessible color choices, and no controls to filter out irrelevant information — making it hard for researchers to extract meaningful insights from the evidence.

Before: The evidence visualization relied on an auto-spinning 3D model with tiny, nearly unclickable targets, creating more distraction than insight and preventing researchers from meaningfully exploring relationships between entities.

Step 1:

The Challenge — Search That Couldn’t Deliver Discovery

When I joined the WDD team, IBM had spent nearly two years building an AI-powered search platform intended to mine over 100 million scientific articles for drug discovery insights — but without any design input or user feedback.

Researchers described the tool as unusable: they couldn’t trust results, interpret visualizations, or understand how the system reached conclusions. Instead of accelerating discovery, the system slowed it down. Usability testing confirmed the problem: the platform failed to support how scientists search, think, and build hypotheses.

Step 2:

Reframing Search Through Deep User Understanding

We embedded ourselves in the daily workflows of scientists at Baylor, MD Anderson, Teva, and Harvard, conducting interviews, lab observations, and usability studies

Key insights reframed the entire product strategy:

Scientists don’t “search” for documents — they explore relationships among genes, diseases, proteins, and pathways.
Literature is the foundation for hypothesis generation, not just fact retrieval.
Trust requires transparency into how results are generated and what evidence supports them.

These insights transformed our design challenge: turn search from a retrieval feature into a discovery engine.

User Research: We completed product testing, user observation and interviews to better understand the domain and microbiologist in universities and industry.

Synthesizing Data:The data from these interviews, observations and product testing was synthesized and organized to form a user’s mental model.

Step 3:

Radically Collaborating to Define a Search-First Product

We convened a two-day design thinking workshop with data scientists, engineers, and PMs to realign the team around these new insights. Together we defined core user outcomes and design principles that guided every decision:

Lower the barrier to entry: use scientific language and workflows to reduce cognitive load.
Support exploration: allow wide-net searches but make it easy to narrow and prioritize.
Promote discovery: reveal relationships and evidence users didn’t know to look for.
Build trust: make system logic transparent and editable.

This alignment laid the foundation for a radically new search experience.

Empathy Maps: During the workshop the teams identified and aligned to the users biggest pain points.

Radical Collaboration: The team came together to define the product release by building empathy for the user and reflecting it on empathy maps, and journey maps. Finally we ideated lots of solutions to solve pain points.

Step 4:

Designing an Intelligent Search Workflow

We rebuilt WDD’s search experience around the way scientists think and work:

Guided faceted search llowed exploration across dozens of structured and unstructured parameters — from genes and biomarkers to pathways, phenotypes, and publication metadata.
Context-rich evidence surfaced how entities were related, why they were surfaced, and where the data originated, including citation metrics and journal impact factors.
Hypothesis iteration enabled users to save, refine, and build on previous searches — shifting from one-off queries to continuous discovery workflows.
Relationship networks visually mapped gene–disease–compound connections, letting scientists explore unknown unknowns rather than just known terms.

This design put scientists back in control, reducing the friction between AI output and scientific reasoning.

As-Is Storyboard: Researchers face relentless pressure to manually sift through endless reports and articles, slowing discovery and increasing cognitive load.

As-Is Storyboard: Current workflows consume weeks compiling and prioritizing findings — time that could be spent advancing research.

To-Be Storyboard: The redesigned experience consolidates all of Watson’s evidence in a single, searchable view, connecting data and relationships in one place.

To-Be Storyboard: By surfacing prioritized candidates instantly, the new workflow dramatically reduces research time and accelerates the path from evidence to experiment.

Step 5:

Delivering Measurable Impact

The results were transformative:

Scientists identified previously unknown gene–disease relationships including ive RNA-binding proteins newly linked to ALS, enabling entirely new research directions
Discovery timelines shrank from months or years to weeks, and research teams described the new workflows as “making the impossible possible.”
The redesigned platform contributed directly to multi-million-dollar enterprise sales on launch day..
Relationship networks visually mapped gene–disease–compound connections, letting scientists explore unknown unknowns rather than just known terms.

“Watson didn’t just speed up our work — it gave us a way to find things we didn’t even know to look for.” – Baylor scientist feedback

Sketch: Early concepts focused on giving researchers a single, comprehensive view of an entity and its relationships — transforming a complex, abstract challenge into a tangible design conversation.

Sketch: We explored ways to simplify the relationship network, including moving advanced options into a dedicated settings drawer to reduce visual complexity and support focused exploration.

Wireframe: As fidelity increased, we designed deeper drill-downs that linked organized information directly to the underlying literature, giving researchers traceable evidence for every connection.

Wireframe: We incorporated accessible color choices and new filtering controls.

Wireframe: Expanded network views revealed how entities were connected.

Transforming Search Into a Discovery Engine

Redesigning Watson for Drug Discovery — accelerating insights from months or years to weeks.

Project Duration:

My Role:

Target User:

Project Type:

Product Team:

Design Team: