AskChem: Claim-Centered Infrastructure for Chemistry Literature Synthesis

The Problem: Chemistry Knowledge Is Fragmented

Scientists and AI systems searching chemistry literature face a fundamental mismatch. When a chemist needs to answer a question like "What solvents work best for this reaction type?", traditional search engines return ranked papers. The researcher must then manually dig through each paper, find relevant passages, verify they're trustworthy, and synthesize answers across multiple sources. This is tedious, error-prone, and doesn't scale—especially when AI agents need to retrieve and cite evidence reliably.

The Claim-Centered Solution

AskChem reimagines retrieval around a smaller, more useful unit: the claim rather than the paper. The authors convert each of 147,000 papers into atomic, typed claims—discrete factual statements extracted and tagged by category. Crucially, every claim carries full provenance: the source DOI and either a verbatim quote or an explicit pointer to where it appears in the paper. This means users and AI agents can immediately verify where a fact came from.

Over this store of 2.4 million grounded claims, AskChem builds three complementary search and discovery structures. First, a stabilized faceted taxonomy—a hierarchical classification system that lets chemists browse by topic (e.g., reaction types, materials, conditions). Second, an evidence graph that links claims through semantic relationships, so users can follow chains of reasoning or conflicting findings across papers. Third, a living taxonomy that situates papers under scientific principles, making the collection explorable from first principles rather than just keyword matches.

The system exposes access through a web interface, REST API, SDK, and MCP (Model Context Protocol—a standard for AI agents), so both human researchers and language models can use it programmatically.

Why It Matters: Grounding and Reliability

The results demonstrate real impact. On a benchmark called AskChem-Bench, when GPT-4.5 used AskChem for retrieval, it produced claims with 100% resolvable citations (you can actually click the DOI and find the paper). Without AskChem, only 88.3% of the model's claims could be verified. Among five tested retrieval systems, AskChem achieved the highest citation density—the number of unique, relevant sources cited per query—meaning it surfaces more diverse evidence rather than recycling the same papers.

For chemistry in particular, this matters. Reactions, materials, and protocols have nuances that depend on context. A claim-centered index makes it possible for an AI system to say "X works under conditions Y, as shown in papers A and B" rather than guessing which paper to cite. It also lets chemists quickly compare what different papers say about the same phenomenon.

The infrastructure is live and currently open, making it a practical tool rather than a research prototype.