Roadmap
From published research to a platform teams can rely on.
We build in the order the evidence allows: methods first, then open tools and pilots, then the platform.
Now
Research foundation
- Mask-aware explainability benchmark for diabetic retinopathy, published with IEEE (2026)
- MLOps foundation: continuous automation for ML, published in Information (2020)
- Toxicity Guard prototype: a reinforcement-learning controller that decides when a language model should answer, clarify, or refuse (2026)
- Steering account for language and vision-language models, manuscript in preparation
- Platform architecture and service repositories
Next
Open tools and first pilots
- Open-source toolkit for scoring explanations, released from the Sight methods
- Lesion-level validation of retinal explanations
- STEER cards and a strong-baseline evaluation harness for steering
- First pilot with a clinical or industry design partner
Later
The platform
- Decision reports that pair each prediction with checked explanations
- Runtime intervention with cost accounting
- Production monitoring: explanation drift, fairness, audit trail
- Evidence packs for high-risk AI compliance
DoneIn progressPlanned
What comes next
Turning methods into tools.
The research exists. The next step is the engineering and partnerships that turn it into something teams can use.
Engineering
Research engineers to turn the published methods into an open toolkit and the first platform services.
Compute
GPU time for steering experiments across open language and vision-language models.
Partnerships
Clinical and industry pilots that test the methods on real models and real decisions.
Help us move to the next phase.
We welcome conversations with investors and design partners.