OHRQPhysical intelligence for engineering and manufacturing

Physical intelligence, measured.

OHRQ is one platform where AI learns real engineering and manufacturing work, and where manufacturers put it to work under their own approval.

A long industrial hall with a glazed roof, work bays down both sides and a track along the middle.
OHRQ BenchPhotograph: illustrative
A person approves every change.Two tracks at release 0.1

For AI labs and manufacturersPhotograph: illustrative--:--:--Z

For AI labs

Benchmarks, evaluation, environments, data.

Benchmarks with deterministic graders, held-out evaluation, training environments and checked data, all in physical engineering.

Hard tasks with objective graders, test sets a model cannot have seen, environments to train in, and data whose labels are checked.

  • Benchmarks
    Versioned benchmark releases with measured floors and ceilings
  • Evals
    Evaluation server and batch grading on held-out tasks
  • Gym
    Reinforcement-learning environments and trainer adapters
  • Data
    Synthetic data products where every row is re-checked by a grader or solver

For manufacturers

Factory system, assistant, engines.

A factory system with an assistant and engineering engines. Nothing changes without a person’s approval.

Answers you can check, plans that fit your ERP, and an assistant that cannot act without a person’s approval.

  • Plantline (factory system)
    Plant data, an assistant and an approvals inbox; imports an ERP’s data and writes an accepted plan back
  • Learn
    Traces, datasets, training, deployment
  • Improve
    Self-improvement loop where every change needs a signed-in person’s approval
  • Engineering engines
    Engineering engines whose answers can be checked. How the platform fits together

Under both: Cloud. Sandboxes, services and simulated computers.

Two customers.One loop between them.

Each side supplies what the other lacks. Labs lack real, under-specified engineering work and real experts. Factories lack models that have been measured on that kind of work. A person approves every change.

1 / 3

Real work becomes tasks and graders.

Real work, opted in and rights-cleared, becomes tasks, graders and data.

2 / 3

Tasks train and evaluate models.

Those tasks, graders and data train and evaluate models.

3 / 3

Better models go back to work.

Better models go back to work and produce more of it. A person approves every change.

Every number comes from a computation,never from a language model.

The same engines serve as tool, grader and data generator.

OHRQ Bench: nine tracks.Two at release 0.1.

Nine tracks from CAD and G-code to plant operations. Every answer is graded by a deterministic grader or a solver, never by a model. Two tracks are at release 0.1, with floors and ceilings measured.

TrackWhat it testsRelease
EngineeringDesign and manufacturing work with one checkable answer: CAD programs, G-code, PLC programs, wiring, tolerances, schedules, inspection plansRelease 0.1
Computer UseOffice and engineering-adjacent work on simulated computers; runs replay to the same stateRelease 0.1
CAD DesktopThe same part-modelling tasks, solved by driving real CAD software from screenshots
Factory AssistantAn assistant inside a running plant system, graded on outcomes
PhysicsPredict stress, temperature and surface pressure on shapes the model has not seen
ProcessMass and energy balances, heat integration, pump and pipe sizing, equipment cost
OperationsScheduling and inventory decisions in a simulated plant
DesignFrom a requirement to a part a solver and the manufacturing rules both accept
Real WorkTasks from real engineering work at companies that opted in

Request access to the benchmarks

Synthetic data,every row re-checked.

Seven synthetic data products, each with a card written from measured numbers and a free sample pack.

  • Checked rows. Every row is re-checked by its source’s grader or solver in a fresh process before release.
  • No overlap with test sets. Anything matching a hidden benchmark task or a held-out engine task is dropped, so sold data cannot contaminate an evaluation.
  • Traceable deliveries. Deliveries use signed, expiring links, and each order can be traced if a copy leaks.

The data catalogue

eng-sftEngineering tasks with checked answers, for supervised training
eng-prefPreference pairs over engineering answers
eng-trajStep-by-step trajectories on engineering tasks
cu-transcriptsComputer-use episodes on simulated computers
proc-qaProcess-engineering questions with computed answers
phys-recordsPhysics records labelled by a solver
mixed-curriculumA curriculum drawn from the products above

Request access to the data catalogue

Measured,on recorded runs.

Frontier agents already pass 32 of 35 synthetic engineering tasks in pilot runs, which is why synthetic-only benchmarks saturate and why OHRQ builds on real work. Every number here comes from OHRQ’s own recorded runs.