RSN Show and Tell · Generative AI

AI for Research Intelligence

Building mental models for research support
Yuyang Wang Research Infrastructure and Project Manager · ICMS, TU/e
1 · Why

Everyone I meet carries a different mental model.

Every day I meet people in many roles. Next to technical ideas, we share research intelligence: who works where, with whom, on what. It is about geography and networks, and each person sorts it their own way.

Research support is managing these mental models, and searching for clarity in a complex environment that changes every week.

2 · How · the human need

Six questions people ask about self-driving labs.

Which labs
names and initiatives
The people behind them
leads and teams
Their domain
materials, chemistry, biology
The tools they master
robots, instruments, AI
Where they are
city, country, region
How they are funded
programmes and budgets
And then people do the maths in their heads

Which city and economy is doing better, and which is falling behind?

Where is this heading? Projections made in the mind, or in an Excel sheet.

2 · How · the journey

From a list by hand to 118 initiatives on a map.

Here the AI-assisted work begins
1

A database by hand

Lab names, people, places and funding, collected from talks, papers and conversations.

2

AI agents search the web

Systematic searches country by country, programme by programme, with a source for every find.

3

I check every entry

Each claim traced to its source before it goes in. People make the final call.

4

One record per lab

The six questions become fields: name, people, domain, tools, place, funding.

5

Map, list and timeline

One data file feeds a map, a list and a timeline, in four languages.

6

Weekly watch

New developments each week, and submissions from the community itself.

3 · What

The global self-driving lab map, live.

0
initiatives worldwide
0
countries
0
community submissions
Andy Cooper
Alán Aspuru-Guzik
shared it and wrote in
3 · What I gain

A research intelligence engine, always ready.

  • Global SDL map · 118 initiatives
  • Literature knowledge base · 845 papers
  • Monthly field radar · new work, every month
  • Deep-research briefs · a group, a topic, a country
  • Policy and funding watch · RAISE, MaterialsCommons

Always ready

An up-to-date answer before every meeting, call or proposal.

A mental model in itself

One shared picture of who does what, where, and with which means.

Clarity for everyone

For me first, and then for every stakeholder I work with.

4 · Reflect

Four lessons from building the map.

1

AI scouts, people verify

Agents find far more than one person can. Every entry still passes a human check.

2

Every claim keeps its source

A link on each fact makes the map citable and easy to correct.

3

Choose the tool by the data

Public data in any tool. Internal data in TU/e on-premise models.

4

Start small, add every week

Begin with a list by hand, and improve it a little every week.

Thank you

Questions?

Backup · How it is built

How the map and the knowledge base are built.

Backup · Data and tools

Sort the data first, then choose the tool.

Public

Open to any tool

  • The map and its sources
  • Published papers and policy texts
  • Public calls and announcements
Internal

TU/e on-premise models

  • Meeting notes and working drafts
  • TU/e on-premise models (SPIKE-1) can handle this data
Sensitive

Kept out of shared AI tools

  • Personal and HR information
  • Confidential partner information
  • Anything under a non-disclosure agreement
Backup · For stakeholder presentations

Two websites I present from.

DiscoveryLabNL briefing site

discoverylabs.nl
  • Pillars, stakeholder map and roadmap on one page
  • Opened live in meetings, link sent afterwards
  • Updated the same day when something changes

TU/e Flagship Advanced Materials

tue-flagship-advanced-materials.nl
  • Vision, mission and milestones in one place
  • Shared context for internal audiences
  • Sets out the ambition for AI and lab automation in materials research
AI for Research Intelligence · RSN Show and Tell, 1 October 20261TU/e
1 / 8
← → navigate · N notes · G overview · F offline view