Chembricks Blog

Physics validation, in practice

Notes from the team on new Virtual Lab features, the methods behind the validation layer, and what is coming next.

A head start from physics beats a bigger AI model

Six AI models, one chemistry problem, and a measured compute bill. Giving a model a fast physics calculation to correct beat making it bigger or feeding it more data.

Read article

After the hackathon: 2,307 molecules against an ebolavirus glycoprotein on a desktop machine

A Boltz-2 co-folding screen of 2,307 compounds against the Bundibugyo ebolavirus glycoprotein pocket, run on a local Tenstorrent workstation. What scored, and how much of it you should believe.

Read article

Building the Chembricks epistemic discovery engine

How we turned LLM-assisted science from a single chat into a controlled multi-branch campaign with a durable evidence ledger.

Read article

Benchmarking our models on chemistry picked to be hard

Verbatim accuracy tables and density parity plots for four production models plus redox, on a benchmark built to be hard, with the failure tails left in.

Read article

From proprietary quantum-chemistry data to physics-gated molecular design

How proprietary quantum-chemistry data becomes trained cmbx_ models, a physics-gated design loop, and two worked case studies with honest limits.

Read article

What's new in the Chembricks Virtual Lab

Four end to end demo workflows, IR and 13C NMR spectra prediction, and provenance for every number.

Read article