IIT-Madras Turned 10,000 Papers Into an Alloy Database an LLM Can Search
Researchers used large language models to extract 1,85,000 structured records from more than 10,000 papers, building a searchable foundation for sustainable alloy discovery. Both databases are free.
Researchers at IIT-Madras have built an AI platform to speed up the discovery of high-performance, sustainable metallic alloys for electric vehicles, aerospace, renewable energy and marine infrastructure.
The method is the story. They used large language models to extract and organise decades of scientific knowledge from more than 10,000 research papers, producing two databases holding over 1,85,000 structured records. Both are freely available, through Alloy Tattvasar and GitHub.
The problem being solved
Materials research has a retrieval problem more than a knowledge problem. The experimental information already exists — it is just scattered across papers, buried in tables, and sometimes only present in figures. Comparing two alloys systematically means a researcher manually reconciling data reported a dozen different ways across a dozen publications.
The automated pipeline pulls out alloy compositions, manufacturing processes, testing conditions, and more than 350 material properties, which makes systematic comparison possible.
"Instead of spending years manually collecting data from thousands of publications, our framework automatically builds structured databases that can be used to identify sustainable materials much faster," said Rohit Batra, assistant professor in the Department of Metallurgical and Materials Engineering.
The extraction technique
The system uses retrieval-augmented generation to pull relevant examples during extraction, which improves accuracy on information drawn from scientific text and tables.
That detail matters more than it might appear. Naive LLM extraction from scientific tables is unreliable in ways that are hard to detect — the output looks plausible and is quietly wrong. Grounding each extraction in retrieved examples is a direct response to that failure mode.
Sustainability is built into the scoring
The platform incorporates environmental, economic and social indicators, so alloys can be assessed on sustainability alongside performance. That is unusual. Most materials databases optimise for properties and leave the sustainability assessment as a separate exercise conducted much later, usually by someone else.
What they demonstrated
The team used the platform to identify promising candidates in three areas: high-entropy alloys for lightweight structural use in automobiles and aerospace, soft magnetic materials for electric motors and transformers, and corrosion-resistant alloys for marine infrastructure, offshore engineering, chemical processing and energy systems.
Why the open release counts
Releasing 1,85,000 structured records publicly, through both a dedicated platform and GitHub, does more for Indian materials research than the discovery results themselves. A structured, searchable corpus is infrastructure. Every group that would otherwise rebuild it from scratch now starts further along.
Source: IIT-Madras AI platform to fast-track discovery of greener, high-performance alloys — DT Next