Scientific knowledge platform that captures unpublished experimental data, structures it into a queryable research record and supports materials-discovery workflows.
Product sources: 83 Sciences official platform site + Y Combinator S26 company page
83 Sciences builds a materials-discovery platform around experimental data that normally remains fragmented or unpublished. Dalton AI ingests historical runs, failed batches, notebooks and instrument exports into a structured, queryable corpus that researchers can interrogate for patterns, anomalies and candidate follow-up experiments. 1 2
AiToolMap did not ingest lab data, reproduce a proposed experiment, validate Dalton's scientific reasoning, inspect a manuscript or confirm a material discovery. No eligible fixed-panel numeric evidence was found. The resulting 7.8 rating is LOW-confidence and not rank eligible.
Scientific recommendations should remain hypotheses to test, not facts. Labs should preserve provenance from raw instrument data through model-generated insight, separate measured observations from inferred relationships, require expert review before physical experiments and protect unpublished IP, patent-sensitive data and partner confidentiality through clear access and retention controls.
The problem selection is strong. Scientific organizations accumulate large amounts of negative and routine experimental data that rarely reaches papers but can still constrain hypotheses and explain why prior attempts failed. Structuring that history into a reusable lab memory is a credible application of AI even before making stronger autonomous-discovery claims. 2
The product surface is unusually domain-specific. Dalton examples reason over temperatures, capacity measurements, XRD anomalies and proposed reruns rather than offering a generic document chatbot. The platform is designed to ingest instrument exports and experimental records directly, which should preserve more scientific structure than a plain RAG layer over publications. 2
Team credibility is also strong: the founders combine chemical engineering, experimental chemistry/ML and enterprise AI backgrounds, with advisors from materials science and AI. Y Combinator independently corroborates the active S26 company and early work with university lab partners. 3 4
Scientific efficacy is not yet independently established. The company says more than 85% of experimental data is discarded, estimates more than $100B in R&D value is lost and says it has turned a partner's discarded data into a co-authored manuscript in under two months. YC also reports a claimed novel-material discovery. These are useful company milestones, but AiToolMap did not find independent peer-reviewed validation showing repeatable discovery lift across labs or materials domains. 1 4
Pricing is partnership/demo-led and no current public numeric price was verified. That limits value comparison and also means the product appears better suited to research organizations with substantial historical datasets than to individual researchers.
Products matched by shared tasks, audience, workflow, product type and capabilities.