BioNTech ·AgentOlympus ·Product Design

Designing AI for experts

AgentOlympus is BioNTech's internal AI platform. I design its specialist workflows and the shared system that connects them.

As the sole product designer, I lead each workflow from problem framing through interaction design and validation, partnering with product managers, engineers, leadership and two or three subject-matter experts per use case.

0
workflows shippedacross five business teams
0
business teamscommercial, market access, medical affairs and more
60+
on the roadmapwhat the system must support
2–3
SMEs per workflowfrom discovery through release

Decoding Data

RoleSole Product Designer
PartnersPMs · Engineers · SMEs
TimelineSince January 2026
ScopeFive business areas
01 · The problem

More time assembling evidence than assessing it.

Across commercial, medical affairs, market research and market access, specialists were doing the same underlying work: gathering evidence from several sources, cross-checking it and turning it into an analysis, recommendation or deck.

Too much time went into assembling and reformatting information. The opportunity was to let AI handle that preparation while experts retained judgement and final review.

Before AgentOlympus
Clinical literature
Pricing data
Congress sessions
Read, cross-checked and reformatted by hand, across several tools
AnalysisRecommendationSlide deck
Manual assembly across several tools
agentolympus
The AgentOlympus home screen, showing the workflow launcher
AgentOlympus

One way in, whatever the work behind it.

Every team starts from the same place.
02 · The platform

Different teams, the same four jobs.

The subject matter changes, but each workflow combines the same four jobs.

Find

Question an evidence base directly.

Compare

Surface differences across sources or markets.

Synthesise

Turn evidence into structured findings.

Produce

Move findings into reports and decks.

What we changed

We introduced an AI layer that retrieves, structures and synthesises evidence while keeping expert review inside the workflow.

AI prepares the material; specialists decide what is trustworthy and what moves forward.

With AgentOlympus
Clinical literature
Pricing data
Congress sessions
The platform assembles
FindCompareSynthesiseProduce
The expert reviews, corrects and decides
AnalysisRecommendationSlide deck
The assembly is handled, the judgement stays with the person

One platform works because the jobs underneath are shared, even when the subject matter is not.

What shipped, and in what order

Each release established something the next workflow could reuse.

  1. Step 01 Histology Data Extraction Dossier Mapping Market Research · HEOR

    Established reliable evidence extraction.

  2. Step 02 OncoAthena Commercial

    Established conversational search and citation.

  3. Step 03 SlideGenie Commercial

    Extended findings into presentation-ready output.

  4. Step 04 TPP Builder Commercial

    Introduced the reusable assessment pattern.

  5. Step 05 Congress AI Medical Affairs

    Brought contributed material into the evidence base for review during medical conferences.

  6. Just launched MR Echo Launched Landscape Assessment Launched

    Extended the conversation and assessment patterns already in place.

HEOR: health economics and outcomes research. GVAPP: global value, access, pricing and policy.

03 · What research changed

Three findings that changed the product

Each workflow is shaped with two or three subject-matter experts. Three findings changed how the platform works.

Finding 01 · Landscape Assessment

Lead with the structured data

Market access specialists already knew which categories mattered. Landscape Assessment now opens with those categories populated for the selected geography, with chat reserved for questions the structure cannot anticipate.

Result

The method lives in the product, not in the prompt.

Landscape Assessment treatment guidelines compared across USA, EU and UK
Easy access. Guidelines, pricing, competition and HTA sit one click apart, already filled in.
MR Echo showing pre-populated Key Business Questions
Pre-loaded questions. Teams begin at a first answer and refine it.
Finding 02 · MR Echo

Pre-load the questions people always ask

Teams repeatedly entered the same Key Business Questions. MR Echo now pre-loads them, allowing people to begin with a first answer and refine it.

Result

Domain knowledge sits in the product instead of relying on prompt skill.

Finding 03 · Congress AI

Put review inside the workflow

Congress attendees contribute shared material, but experts warned that not every upload should become evidence. Content now remains in review until the team approves it for AI use.

Result

Governance becomes a product behaviour, not a separate policy.

Congress AI dashboard showing documents awaiting review
Review before trust. Uploads wait in Pending review until the team lets them in.
04 · Designing for scale

Designing a system around workflows

With more than sixty workflows on the roadmap, designing each independently would not scale. I am building shared foundations, interaction rules and configurable patterns alongside the products.

Citation is central: specialists need to inspect the source before acting on an AI answer.

01
Foundations
Colour, type, spacing and base components.
02
Shared patterns
Layouts and behaviours reused across workflows.
03
Workflow configuration
What a workflow changes to suit its own job.
From shared to specific

Getting to a design language first

None of that is possible until there is a design language leadership recognises as BioNTech's. Agreeing one took longer than drawing any single screen.

The blocker

The direction existed, but only in conversation.

Leadership's preferences existed, but had not been translated into shared principles.

The decision

Start from what is settled, then offer a choice.

I separated established brand decisions from areas that still required alignment.

What changed it

Clickable beat flat, every time.

Clickable directions produced clearer decisions than static screens.

01 Foundations

Colour, type and the base components every workflow inherits.

Six colour ramps, a Figtree type scale with Source Serif 4 reserved for AI-written text, 4pt spacing and an eight-step radius scale now support the platform's shared components.

AgentOlympus design systemFigtree + Source Serif 4 · 6 ramps · 4pt spacing
Colour
Jungle GreenBrand primary#005A64
Forest GreenBrand secondary#3C9673
Lime GreenBrand accent#BECE33
Space CrayolaNeutral#294145
SinopiaError#C73F1E
MarigoldWarning#FFA62B
Type
StyleWeightSizeLineTracking H1Medium5003240-1 H2SemiBold6002432-0.5 H3SemiBold60020240 H4SemiBold60016240 Body LRegular40016240 Body MRegular40014200 Body SRegular40012160 AI bodySource Serif 440014200
Components
Buttons
Labels
NeutralBrandSuccessWarningErrorInfo
Corner radius
NoneXSSMLXL2XLRound
Spacing · 4 to 80
02 Shared patterns

Two patterns carry most of the platform.

Most workflows reuse two arrangements: conversation and assessment.

Conversation Scoped chat on the left, a source panel on the right, citations that open into the evidence.
OncoAthena chat with a SharePoint source tree beside it
OncoAthena. The source panel is a document tree.
MR Echo citation panel with an interview transcript and the cited passage highlighted
MR Echo. The same panel holds a transcript and the cited moment.
Assessment Filter chips over a comparison table, with the same row, header and expand behaviour.
TPP Builder standard of care and emerging benchmark tables
TPP Builder. Standard of care against emerging benchmarks.
Landscape Assessment price benchmarks across USA, Germany and UK
Landscape Assessment. The same table, pricing across geographies.
03 Workflow configuration

And the room a workflow needs to be itself.

Workflows can depart from the two dominant patterns without leaving the system. SlideGenie uses the same foundations for a workflow that begins with a prompt and ends in a deck.

SlideGenie generating a presentation from a prompt
SlideGenie. Starts from a prompt, ends in a deck on BioNTech templates.
Landscape Assessment competitive timeline of approvals, market entry and exclusivity
Landscape Assessment. The same system stretched to a timeline of approvals and exclusivity.

The goalOne design language, configured to each workflow.

05 · Where it stands

Adoption and current evidence

Eight workflows are shipped across five business teams, including commercial, market access and medical affairs. The platform grew from a two-workflow pilot with fifty oncology specialists to more than two hundred users across the organisation.

When the same need appears twice, it becomes a shared pattern instead of another one-off solution.

200+Users across five business teams, up from a pilot of fifty, with Congress AI running during live conferences to derive insights and summaries
60%Faster to launch, from eight weeks to about three: AI-built prototypes handed to engineering, four workflows built in parallel on the shared design system
50%Less time on evidence synthesis in pilot testing, with source grounding, citations, confidence scoring and expert correction built into the workflow
The AgentOlympus launcher, listing every agent workflow with its business area
The platform today. Every workflow reachable from one place, labelled by the business area it serves.
Takeaways

Designing AI is more than a chatbot.

The work is identifying what AI should handle and preserving enough expert control for the result to remain trustworthy.

A platform becomes one product when shared patterns support distinctly different work.

Next project Breaking Money Silos