Speeding Up Merck’s Process from 6 Months to 6 Hours with an AI R&D Assistant
6
Hours to find molecules
vs. 6 months of manual work by domain experts.
5
Weeks to build it
The window for the proof of concept, met and under budget.
65
Countries
Where Merck operates, making the tool relevant across its global business.
The client
Merck is a German science and technology company working across healthcare, life science and electronics, with around 62,000 employees in 65 countries.
With roots going back to 1668, it is the world's oldest operating chemical and pharmaceutical company. Its domain experts track new chemical compounds across a constant flow of scientific literature.
The challengeWhat the teams actually needed
Merck's domain experts spent around 6 months reading scientific literature by hand to find chemical compounds worth selling. Merck had 5 weeks to see whether AI could take that work over.
Challenge 01Input bigger than the model
The volume of literature ran past what popular LLMs can hold, so the input had to be broken into chunks before anything could read it.
Challenge 02A catalog missing key data
Merck's internal catalog lacked details like InChIKeys and CAS numbers, so matching found compounds to products wasn't possible.
Challenge 03Five weeks, no more
The proof of concept had to be ready inside the deadline and inside budget.
What we did
An AI R&D assistant, built in 5 weeks
We built an AI R&D assistant that turns scientific papers into a ready list of chemical compounds for Merck's domain experts.
Chemicals named and coded
Every chemical mentioned in an article is identified along with its role in the system, then given its official name and unique code so there is no ambiguity about what was found.
Properties from the big databases
Large chemical databases supply the rest: InchiKey, Smiles code, molecular formula, synonyms, and CAS numbers pulled from those synonyms.
Already on the shelf?
Each identified molecule is checked against the Merck catalogue database to see whether the company already sells it.
Results a chemist can read
Domain experts upload PDFs and see all results on one screen, including a 2D image of each chemical structure.
Client testimonial
If I had to choose two words to summarize our collaboration with Netguru it would be speed and efficiency. They were able to provide us with a POC within the 5 weeks deadline and now, with the AI R&D Assistant, we can complete our task in a fraction of the time.
Tech stackThe tools that made it possible
01
LangChain
The framework behind the extractor, calling out to the model endpoints.
02
Azure OpenAI
The endpoints doing the extraction work.
03
AWS
Merck’s own secure infrastructure, where the proof of concept is hosted.
04
Chemical databases
External sources supplying the properties behind each identified molecule.
More like thisOther places we scaled product teams for growth.
What’s next