Drylab Blog

Animation on dark green: without the selector, a request reaches six tools and the agent picks a wrong one, with a card listing why; with the selector, the whole catalog lights up, a line reaches boltz2 and its tool card opens.

Featured

A scientific agent that chooses the right tools

Our scientific agent can use a wide range of tools, pipelines, databases, analysis operations and method guides, but in production it often misses the right one. We built the first benchmark measuring ability to discover and select the right tool for scientific research tasks. Later on we trained a Jev-like selector that outperforms Claude Code, Codex and Biomni on the same set of resources.

Technical13 min read
Read the post

All posts

5 posts · featured post above

Illustration: a stack of atlas files flowing into single-cell clusters and layered spatial slides, on dark green
Playbook11 min read

Omics at atlas scale: rebuilding single-cell and spatial bioinformatics infrastructure for the agentic era

In this post, we describe how we rebuilt the single-cell and spatial analysis methods our agents depend on so that they keep working at atlas scale. Using a multi-agent system, we rewrote these methods for GPUs, plus a few high-performance CPU and Rust ports, deployed them on scalable serverless infrastructure, checked each one against the original tool, and gave our agents access to all of them through a single API.

Illustration: branching white lines with nodes on dark green, like candidate routes splitting from one path
Technical24 min read

DryLab Discovery at the OpenADMET PXR challenge

In this post, we share a technical post-mortem of how our autonomy research system automated feature engineering, diagnosed validation-loop biases, and achieved a competitive, near-state-of-the-art result on entirely novel chemical scaffolds.