Real-world data for models that do real work

Organizational datasets, engineering CAD corpora and RL environments, built by the people who do the work.

Data that can't be scraped.
Environments that can be verified.

Organizational data

CAD & engineering

RL environments

Licensed.Consented sources.
Expert-graded tasks.
Rewards you can verify in code.

Three data products. One lab.

Organizational data

How companies actually work

De-identified org charts, tickets, email threads, docs and approvals from consenting partner companies. Linked across systems, so models learn the hand-offs, not just the documents.

Learn more

CAD data

Parametric engineering corpora

Native CAD with full feature history, sketches, constraints, assemblies and drawings, annotated with design intent by professional mechanical engineers.

Learn more

RL environments

Environments with checkable rewards

Containerized replicas of enterprise software and CAD workflows. Seeded state, long-horizon tasks and programmatic graders your trainer can call millions of times.

Learn more

Evaluation & QA

Graded by domain experts

Every task reviewed by practitioners, from ops leads to senior engineers. Inter-rater agreement, reward-hacking audits and held-out splits ship with every delivery.

Learn more

Request the Dataset Evaluation Package.

Contact Sales

Delivered in the formats your training stack expects

From raw CAD files to ready-to-run environment containers.

STEP · B-repnative CAD + feature tree
JSONLthreads, traces, rubrics
Gym APIreset() · step() · reward
OCI imagescontainerized environments

Main advantages

The reasons frontier teams choose and trust Glasa Lab.

Clean provenanceLicensed from consenting organizations and paid experts, with full chain of custody.
Rewards you can trustProgrammatic graders with adversarial audits, so policies learn the task, not the grader.
Native CAD, not meshesFeature trees, sketches and constraints preserved, with the design intent behind them.
Real organizational contextCross-system threads linking people, roles and decisions.
Built to scaleContainerized environments that parallelize across your cluster, with fresh splits every month.

Need a custom environment?

Scope a Build