Work & expertise
What we can show you.
Client work in regulated industries is mostly confidential. The research, frameworks, and standards work behind it are public and citable, so that is what we show.
Industries
Most of our work happens under a regulator.
- Healthcare & PharmaOur anchor. A decade building AI that runs in hospitals: imaging pipelines, multi-site collaboration without data sharing, validation that satisfies clinical review.
- Academia & ResearchUniversities, research hospitals, and consortia. We turn grant-funded prototypes into sustainable, validated software, and carry research models the last mile into clinical deployment.
- Financial ServicesDocument intelligence, risk models, and LLM workflows, deployed with the lineage and audit trail your regulators expect.
- Energy & UtilitiesTransmission developers, utilities, and the grid that cannot go down. We unlock capacity on the lines already in the ground: dynamic line rating, topology optimization, and sovereign siting and permitting copilots that never leave your perimeter.
- ManufacturingDefect detection and predictive maintenance that hold up at line speed, on the hardware you already run.
- AutomotivePerception and analytics built under hard latency, safety, and certification constraints.
- HospitalityHotel groups and multi-property operators. We build the data spine under the banquet hall, the restaurants, and the spa, and hold the guest file to the standard a data regulator now expects.
Representative engagements
The engagements we run most often.
Client names stay confidential, so what follows are shapes of work: each one an engagement we deliver end to end, from architecture through monitoring.
The pilot that stalled before production
The model that wowed in a demo and then stalled, or the prototype you inherited and can’t trust. We re-engineer it into something your ops team can run and your auditors can sign off: hardened pipelines, lineage, monitoring, drift detection, and the tests that got skipped.
Research model → clinical deployment
Take a promising model from the lab to validated deployment on hospital infrastructure: data harmonization, retraining, and inference optimization for the hardware you actually have.
Learning across sites, without moving data
Stand up federated training across hospitals, partners, or geographies, so consortia can build shared models while every record stays home.
Private LLMs on your data
Adapt open or frontier language models to your domain inside your own environment: fine-tuning, evaluation harnesses, and guardrails included.
Document intelligence for regulated back offices
Extraction, classification, and review workflows for finance and compliance teams, with every model decision logged and reviewable.
AI at the edge of the plant
Defect detection and predictive maintenance under real latency, bandwidth, and hardware budgets: production lines, vehicles, substations.
The AI cost-down audit
A focused pass over your training and inference spend. Our track record: training costs cut by up to 50%, inference latency by up to 70%, without giving up accuracy.2
Built in the open
Open-source infrastructure, used in production.
- GaNDLFLow-code, reproducible deep learning for clinical workflows.Communications Engineering (Nature) Editor’s Choice · an MLCommons project · Created and led by our founderGitHub ↗
- MedPerfFederated benchmarking of medical AI at global scale.Nature Machine Intelligence · an MLCommons project · Core teamGitHub ↗
- FeTSReal-world federated tumor segmentation across 71 sites on 6 continents.Nature Communications · covered by The Wall Street Journal · Co-led by our founderGitHub ↗
- OpenFLAn open framework for federated learning, hardened in healthcare.Physics in Medicine & Biology · Core contributorGitHub ↗
- CaPTkQuantitative cancer-imaging platform for radiomics and ML phenotyping.Journal of Medical Imaging · Co-led by our founderGitHub ↗
- GaNDLF-SynthDemocratizing generative AI for medical imaging: autoencoders to diffusion.MLCommons ecosystem · Created by our founderGitHub ↗
Plus 40+ conda-forge packages maintained for reproducible scientific computing.
Selected publications
Six papers we’d point a reviewer at.
- Federated learning enables big data for rare cancer boundary detection
Nature Communications · 2022 · 71 sites, 6 continents: the largest real-world federated learning study to date
- GaNDLF: a generally nuanced deep learning framework for scalable end-to-end clinical workflows
Communications Engineering (Nature) · 2023 · Editor’s Choice
- Federated benchmarking of medical artificial intelligence with MedPerf
Nature Machine Intelligence · 2023
- OpenFL: the open federated learning library
Physics in Medicine & Biology · 2022
- Towards fair decentralized benchmarking of healthcare AI with the FeTS Challenge
Nature Communications · 2025
- Privacy preservation for federated learning in health care
Patterns (Cell Press) · 2024
Standards & community
We help write the rules we build by.
- Vice Chair, Algorithmic Development, MLCommons Medical Working Group
- Organizers of leading benchmarks: the BraTS and FeTS challenges (MICCAI)
- Contributors to imaging standards: IBSI and AI-RANO
- Reviewers for Nature Communications, IEEE Transactions on Medical Imaging, and Radiology
- Tutorial faculty at MICCAI, AAAI, ISBI, and RSNA
Tell us what you’re trying to ship.
Most of this starts with a 30-minute call. If there’s something there, the usual next step is a 2–4 week discovery sprint: framing, feasibility, and a costed plan you keep either way.
Sources & attribution
- 2. Training-cost reductions delivered in Vaiyu client engagements; latency and resource reductions delivered by our founder at Indiana University.