Home
AI · MVP Design · B2B SaaS

Simplifying Scientific Research with AI-Powered Target Discovery

Shaped the UX direction for an AI-powered research MVP that transformed how scientists explore data, assess targets, and create documents.

Knowledge Graphs Complex Workflows AI-based Features Benchmark Analysis UX Strategy Information Architecture Wireframes Design Interaction Design
↑
Boosted user engagement
↑
Faster document creation
01
Product Overview

Where AI Meets Scientific Discovery

SciGenie is an AI-powered research platform. It helps researchers explore disease-gene links, verify scientific evidence, and create documents more efficiently. The platform brings together scientific literature, competitor insights, internal research, and teamwork in one shared space. This allows teams to go from exploring ideas to final scientific proposals faster.

To comply with my non-disclosure agreement, I have omitted and obfuscated confidential information in this case study. All information in this case study is my own.

02
Rationale Behind the Goal

Connecting research, evaluation, and collaboration

Researchers needed to switch between many different tools to explore disease-gene links, review evidence, and create target proposals. The process was slow, fragmented, and hard to coordinate across teams.

  • User Tool chaos Switching between systems to gather information.
  • User Manual synthesis Proposal creation took heavy hand-work.
  • User Unclear evaluation Criteria and decision steps were unclear.
  • User Weak collaboration Hard to share knowledge across teams.
The Goal

One AI-assisted platform for research, proposals, evaluation and collaboration — making the process faster, clearer and more team-aligned.

  • Business Missed targets Goals not met for several years.
  • Business Too few proposals Pipeline thin; innovation slowed.
  • Business Knowledge silos Internal research hard to reuse.
  • Business Slow decisions Outdated info weakened confidence.
03
Users & Audience

Two Sides of Scientific Decision-Making

SciGenie supported two closely connected groups: researchers exploring scientific opportunities and evaluators who reviewed and validated proposals. They aimed for the same goal, but their workflows, priorities, and information needs varied greatly.

Workflow: Researchers handle Exploration and Target Proposal Creation; Evaluators handle Evaluation, Peer Review, and Decision. Evaluation and Peer Review loop feedback back to Target Proposal Creation.

Researchers

01 Exploration
02 Target Proposal Creation

Evaluators

03 Evaluation
04 Peer Review
05 Decision
Group 01

Researchers

Focus
Discover promising therapeutic targets
Works with
Peers, literature, patents, internal & competitor data
Output
Target proposals submitted for review

Researchers explored disease–gene relationships by navigating scientific literature, patents, internal projects, competitor insights, and conference updates while collaborating with their peers to identify promising therapeutic targets and create proposals.

Their workflow focused on

  • Exploring disease–gene associations
  • Reviewing literature, patents, and latest scientific news
  • Accessing internal historical research and competitor insights
  • Collaborating with peers in shared workspaces
  • Generating and editing AI-assisted target proposals
  • Submitting proposals for scientific review
Group 02

Evaluators

Focus
Validate scientific quality & consistency
Works with
Proposals, supporting evidence, researchers
Output
Reviews, feedback & decisions

Evaluators were experienced researchers and subject matter experts responsible for reviewing and assessing target proposals. Their role focused on ensuring scientific quality, consistency, and alignment across evaluations while providing actionable feedback to researchers.

Their workflow focused on

  • Reviewing target proposals and supporting evidence
  • Assessing scientific validity and proposal quality
  • Conducting peer reviews and providing feedback
  • Comparing proposals against evaluation criteria
  • Collaborating with researchers during review cycles
  • Supporting transparent and consistent decision-making
04
My Role & Collaboration

Designing Alongside Product, Data, and Engineering Teams

I owned the end-to-end product design for the MVP and, in parallel, contributed from the business analysis side — shaping what to build as much as how it should work.

What I owned

Product design

Workflow design Information architecture UX strategy Interaction design UI direction Prototyping Content structure

Business analysis

Requirements Workflow mapping Scope clarity Stakeholder alignment Translating AI complexity

Product direction shifted often. Working closely with Product Owners, engineering, and data teams kept user needs, AI capabilities, technical limits, and delivery pressure in balance.

Who I worked with

Product Manager & Product Owners

Shaped workflows, defined requirements, prioritized MVP scope, and refined future product direction.

Data Engineer & Data Analyst

Clarified AI capabilities, data structures, and the feasibility of exploratory, AI-assisted workflows.

Frontend Developers

Aligned interaction patterns and interface decisions with technical constraints and implementation effort.

Scientific Stakeholders & SMEs

Validated terminology, proposal structures, evaluation criteria, and domain-specific workflows.

Extra contribution Ran usability checks and exploratory testing to surface edge cases, risky interactions, and points of friction — all within tight deadlines.
05
Constraints & Challenges

Designing Around Data and AI Constraints

SciGenie was developed as a fast-moving MVP within evolving AI and data constraints. While the long-term vision was ambitious, the challenge was identifying which workflows could realistically deliver value within four months — while balancing scientific accuracy, technical feasibility, and usability.

01

AI Capabilities vs. User Expectations

The AI workflows needed clear prompts. They couldn’t handle fully open-ended chats like ChatGPT. To improve reliability and ease of use, the interface offered ready-to-use prompt examples and familiar chat patterns.

Key trade-offs

  • Structured prompts instead of unrestricted AI conversations
  • Guided interaction patterns over experimental AI behaviours

02

Fragmented and Incomplete Data

Some key scientific and historical project data was missing, inconsistent, or scattered across different systems. This affected the scope and accuracy of several AI-assisted workflows, especially proposal generation and competitor analysis.

Impact on the MVP

  • Proposal drafts were more generic than initially planned
  • Historical project analysis via AI had to be postponed
  • Competitor insights were partially limited by the available data
  • Only selected therapeutic areas were included in the first release

03

Balancing MVP Simplicity
with Scalability

The original vision included broader collaboration and AI-assisted capabilities. Due to timeline and feasibility constraints, we prioritised delivering the core research and proposal workflows first while intentionally designing foundations for later.

Deferred for future rollout

  • Google Chat and Meet integration
  • AI-assisted proposal history analysis
  • Automatic graph and image generation inside proposals
  • Advanced proposal customisation using additional metadata

04

Limited Time for Validation

The four-month timeline left little room for formal usability testing. Validation relied on lightweight remote feedback sessions, rapid stakeholder reviews, guerrilla testing with user representatives, and some exploratory testing.

Validation approach

  • Quick feedback loops with researchers and evaluators
  • Prototype walkthroughs during sprint reviews and demos
  • Manual edge-case testing and usability checks
  • Iterative refinement alongside development discussions

The real challenge

The biggest challenge wasn’t creating perfect AI workflows (spoiler: “perfect” does not exist). It was about making workflows that were clear, trustworthy, and practical, given the data we had and the changing AI technology.

06
Product Strategy & Direction

Building on Existing Foundations

Unlike a typical greenfield project, SciGenie had existing research findings, early concepts, requirements, and rough mockups when I joined. The challenge wasn't starting from scratch. It was about making sense of fragmented inputs, changing expectations, and technical limits to define a realistic MVP direction. Instead of a strict linear process, the work evolved through ongoing discovery, design, validation, and prioritisation.

Turning Existing Knowledge
Into Actionable Product Decisions

The project had already gone through early discovery activities, but knowledge was spread across research reports, requirements documents, presentations, user stories, and unfinished design concepts. Instead of restarting discovery, I built upon existing knowledge and identified the gaps that still needed validation or refinement.

What I analyzed
User interview notes UX research report Early requirements Existing user stories Pitch & pilot presentations Rough Figma concepts Team comments & feedback Initial workflow definitions
Core problems
  • Information Fragmentation Scientists need to switch between PubMed, Google Patents, and gene databases. This manual process makes their workflow very inefficient.
  • Information Overload Reading hundreds of papers takes a lot of time. It's tough because of conflicting data and figuring out what counts as "true" evidence.
  • Internal Knowledge Gaps Many people don't know about past projects or the current experts in the company. This leads to missed chances to collaborate.
  • Manual Coordination Gathering peer feedback and combining months of research into a final proposal takes time. It can also lead to inconsistencies.
Key recommendations
  • Centralised "One-Stop Shop" Combine literature, patents, and gene expression data into one platform.
  • Intelligent Filtering & AI Use AI to summarise literature. Let users sort research by impact or citations, not just by date.
  • Internal Expertise Mapping Build a database of current and past projects. This will help identify active internal experts for peer review.
  • Transparency & Trust Build confidence by clearly explaining "evidence scores." Use visual indicators, like colour codes, for clinically validated data.
Open questions
  • Visual Utility Is a complex network map actually useful, or would scientists prefer a simple list of links and resources?
  • Defining Competition Should "active competition" be based only on target-disease matches, or can it include any work on a specific gene?
  • Patent Relevance How should patent data be integrated for biological scientists who rarely use it?

Learning From Existing AI
and Research Products

The product introduced AI-assisted scientific exploration at a time when conversational interfaces were rapidly becoming mainstream. I wanted to understand which interaction patterns users would already recognize and trust. The analysis helped establish familiar patterns for AI search, proposal editing, and collaborative research workspaces.

What I benchmarked

A few representative examples

ChatGPT Perplexity Consensus Elicit Google Docs Google Drive Google Workspace Asana Coda

Key questions

What each teardown was looking for

  • →How do users start exploration?
  • →How are prompts guided?
  • →How is information organized?
  • →How do collaborative workspaces work?
  • →How should generated content be edited?
1

Make the first prompt effortless: ready-made examples and easy to use chat

A chat box can be intimidating when users are unsure what a scientific assistant accepts as input. The tools I studied addressed this by showing example prompts and reducing unnecessary interface clutter. Clickable suggestions provide guidance and inspiration, helping users understand what to ask and where to explore next. The large, distraction-free input area encourages the first question, after which the interface transitions into a familiar chat experience.

2

Make results skimmable, then let filters do the narrowing

A literature search is useful only when results are clear and easy to browse. The best examples included detailed entries – authors, journal, year, citation count – but remained skimmable. They used icons, bold titles, and consistent formatting to create a smooth reading rhythm. A rich set of filters, either in a top bar or side panel, helped narrow down long lists without endless scrolling.

3

Unite the team in one shared, collaborative workspace

A workspace is the shared home for everything a research effort produces – documents, findings, and discoveries in one place. The tools I studied organised that content as separate boxes in folder-like workspaces. They showed extra details through a side panel and, most importantly, made sharing easy. That last point is key: a workspace proves its value when it brings researchers together and helps them move faster as a team.

4

Make editing familiar and reviewing effortless

Writing the target-proposal document and running peer review both rely on content that is easy to edit and comment on. The tools I studied used familiar conventions, such as formatting toolbars from Google Docs and Word, eliminating the need to learn new editing patterns. Reviewing was equally lightweight: users could click any item, leave a note, and submit it. Borrowing these established patterns keeps the focus on the science rather than on learning the interface.

How it influenced the design

Each teardown pointed to one principle I could carry straight into SciGenie. Together they cover the product's core moments — starting a search, reading results, working as a team, and shaping the output — by leaning on patterns researchers already trust.

Key design takeaways

01

From the starting-point study

Make the first prompt effortless

A blank chat box intimidates — so ready-made example prompts and a large, distraction-free input invite the first question and turn hesitation into momentum.

02

From the results & filtering study

Make results skimmable, then filterable

Dense literature stays usable when each entry reads at a glance — icons, bold titles, steady formatting — and a rich filter set, including a publication-date timeline, narrows it to what matters.

03

From the workspace study

Make the workspace shared easily

Research moves faster when everything lives in one shared space — content held as tiles, detail surfaced on demand in a side panel, and sharing only a click away.

04

From the content study

Borrow patterns people already know

Editing and review feel effortless when they reuse familiar conventions — a Docs-style formatting toolbar and a click-to-comment flow — so attention stays on the science, not the tool.

In summary

The biggest lesson from this phase was that designing an AI-powered product is about balancing user needs, scientific complexity, available data, and technical constraints. Success came from focusing on the capabilities that could deliver the most value in the first MVP.

07
The Design

Letting Researchers Do the Research...

...and taking care of the rest.

To reduce the burden on researchers and scientists, I envisioned a scientific exploration workspace that would streamline complex workflows and minimize cognitive effort. By bringing fragmented activities into a single tool, creating dedicated spaces for different expert groups, and leveraging AI where it added the most value, the product enabled users to focus on evaluating evidence rather than managing processes.

The design focused on the workflows that created the most value for researchers and evaluators: exploring scientific evidence, organizing findings, generating target proposal documents, and reviewing them collaboratively. The result was a simpler, more efficient experience for tackling highly complex scientific work.

01 AI-Based Scientific Exploration

Lowering the Barrier to Evidence Discovery

Problem

Researchers had to navigate literature, patents, competitor insights, internal projects, and expert knowledge across multiple disconnected sources.

Finding relevant information often meant switching between tools and manually combining findings.

Solution

Introduced an AI-assisted exploration experience that provided a single entry point into multiple knowledge sources.

Key decisions:

  • Chat-based interaction
  • Guided prompt examples
  • Natural language search
  • Unified results experience
  • Familiar AI patterns

Impact: Guided prompt examples gave researchers a clear starting point, removing the blank-page hesitation and helping them reach relevant evidence faster.

Impact: Asking questions in plain language let researchers search across all knowledge sources at once, so they spent their time evaluating findings instead of juggling tools and keywords.

02 Organizing Research in Workspace

Connecting Insights Across Sources and Teams

Problem

Finding information was only part of the job.

Researchers also needed a way to organize discoveries, revisit them later, and collaborate with others.

Solution

Created collaborative workspaces where users could collect relevant findings from different sources and organize them around a target under investigation.

Users could save:

  • literature
  • patents
  • competitor information
  • expert contacts
  • internal projects

… and more, in one shared space.

Impact: Creating a dedicated workspace gave researchers a single home for an investigation, so findings from different sources stayed connected instead of scattered across tools.

Impact: Expandable details let users capture high-level context for each workspace, making it easy to revisit the findings long after the space was created.

Impact: Saving directly into an existing workspace let researchers build on earlier work in place, so related evidence accumulated around the same target over time.

Impact: Sharing a workspace turned individual discovery into team knowledge, letting collaborators pick up an investigation without re-searching the same sources.

03 Generating Target Proposal

Reducing Proposal Creation From Days to Minutes

Problem

Creating target proposals required researchers to manually gather evidence from many sources and structure it into a formal document.

The process was time-consuming and repetitive.

Solution

Introduced AI-assisted proposal generation.

Researchers could:

  • collect evidence
  • generate a proposal draft
  • edit content directly
  • submit for review

… without leaving SciGenie platform.

Impact: One-click generation turned scattered evidence into a structured draft, collapsing days of manual assembly into minutes.

Impact: Easy and intuitive editing let researchers refine the draft in place, keeping full control over the final wording without exporting to another tool.

Impact: Submitting straight from the workspace moved proposals into review without leaving the platform, closing the loop between evidence and decision.

04 Reviewing Target Proposal

Bringing Transparency to the Evaluation Process

Problem

Researchers often struggled to understand how proposals were evaluated and what criteria influenced decisions.

The review process lacked visibility, was not clearly defined, and was scattered across separate tools rather than integrated into one environment.

Solution

Designed a structured review workflow that helped evaluators:

  • assess proposals
  • provide feedback
  • participate in peer reviews
  • communicate decisions

… within the same environment.

Impact: Requesting a proposal peer review directly from SciGenie brought evaluators into the same environment, replacing scattered email threads with a clear, traceable review request.

Impact: Providing a peer review while being able to see the exact review request at all times, made the process context-driven and simple.

Impact: Routing feedback straight back to the author closed the review loop in one place, giving researchers clear, actionable direction without chasing decisions across separate tools.

08
Outcome

Validating a New Approach
to Scientific Research

SciGenie was delivered as a working MVP under significant technical, data, and timeline constraints. While some planned capabilities were intentionally postponed or simplified, the project successfully validated the core product direction and established a foundation for future AI-powered research workflows.

Measured Impact

↑

Boosted user engagement

↑

Faster document creation

Key Benefits

01

Faster Proposal Creation

Researchers could generate proposal drafts in seconds instead of manually compiling information across multiple tools and documents.

02

Clearer Evaluation Process

The redesigned review flow provided greater visibility into how proposals were assessed, helping researchers better understand evaluation criteria and feedback.

03

Centralized Scientific Exploration

Literature, patents, competitor insights, internal projects, and expert knowledge became accessible through a single environment rather than multiple disconnected systems.

04

Better Collaboration

Shared workspaces improved visibility into ongoing research activities and enabled teams to build upon each other's findings more effectively.

A Strategic Win Beyond the MVP

Establishing a Foundation for AI-Assisted Research

The MVP demonstrated how AI can streamline research and proposal creation without taking control away from researchers.

09
Lessons Learned

Lessons From Designing an AI-Powered Research Platform

01

Involve Timeline Owners Earlier in Trade-off Discussions

The project required continuous prioritization as technical constraints and data availability became clearer.

Looking back, earlier conversations about extending timelines or adjusting delivery expectations could have created space for a more complete first version of several high-value capabilities, particularly around proposal generation and collaboration.

If I did it again

I would initiate scope-versus-timeline discussions earlier and more proactively whenever emerging constraints threatened the intended user experience.

02

Not Every Technical Term Should Be Simplified

Scientific domains depend on highly specialized terminology, but not all users share the same level of expertise.

Throughout the project, I frequently worked with Product Owners and subject matter experts to understand which concepts could be simplified and which needed to remain precise.

If I did it again

I would spend more time validating terminology with different user groups to distinguish between language that clarifies and language that confuses.

The biggest takeaway from SciGenie was learning how to design for an emerging technology while operating within real-world constraints. Success did not come from building the most advanced AI experience possible — it came from identifying where AI could provide meaningful value today, designing appropriate guidance around its limitations, and creating a product that researchers could realistically adopt and trust.

Copied