Approach
How we build software people can rely on.
Our products differ, but the method does not. These are the rules every ArcRos system is designed around, and the things we have decided not to do.
Principles
Six rules, applied to every product.
- 01
Ground every claim
A system that produces medical content should be able to show where each statement came from. MedKey attaches a source citation to every question. Cresca shows the evidence behind each option it ranks.
- 02
Treat abstention as an answer
'Not enough evidence' is a valid and useful output. Our systems are built to decline rather than fill a gap with something plausible, and that path is designed and tested like any other feature.
- 03
Keep patient data private by default
Personal health information does not go to a large language model unless a deployment is explicitly configured to allow it. Privacy is a default setting, not an upgrade.
- 04
Offer options, leave the decision
When software recommends, it should show the reasonable alternatives and their tradeoffs. Clinicians and educators hold the judgment and the accountability. The software supports them.
- 05
Gate releases on readiness
In regulated domains, shipping early is not a virtue. A MedKey exam track opens only when it passes its readiness gate. Cresca reaches clinics through planned pilots, not a public launch.
- 06
Keep people who carry the risk in the loop
Faculty review what students see. Clinicians shape what interrupts them. The people accountable for an outcome get a say in the software before it reaches them.
What we will not do
Lines we have already drawn.
- Invent a citation, a source or a statistic.
- Send patient data to a language model by default.
- Present a study tool as a clinical tool.
- Claim approvals, validations or customers we do not have.
- Open an exam track before it passes its readiness gate.
- Announce a defense product before it exists.
Research direction
Spatial computing for training under pressure.
We are exploring AR and VR software for medicine and defense, where practicing a procedure or a scenario in three dimensions could change how people train. This is research. There is no product, and we will hold it to the same rules above before there is one.
Research direction · Not a product
Hold us to these rules.
If you see us fall short of any of them, tell us. It is the most useful feedback we can get.