Thermite for Healthcare and life sciences
Improve pipelines and models without moving patient data.
Genomic, imaging and clinical data often cannot leave your infrastructure, and Thermite does not require it to. Agents improve pipelines and models against truth sets and annotations on your own machines. Goal campaigns take an API through its conformance tests; research campaigns read the trial registries and the literature on a clinical question.
Genomics pipeline acceleration
Agents tune the pipeline's tools, threads and scattering. The judge times it on your servers and scores its calls with hap.py against a truth set.
For Solo builders · Enterprises
- Measures
- Minutes from reads to calls
- Holds
- SNV and indel F1 held
- Hidden
- A sample the agents never see
Medical image segmentation
Agents change the training recipe, the loss, augmentation and post-processing. The judge trains each version on a fixed budget and scores Dice on scans they never see.
For Startups · Enterprises
- Measures
- Mean Dice on a fixed training budget
- Holds
- Each structure's Dice and HD95 held
- Hidden
- Scans from a site never seen
Clinical note de-identification
Agents tune the patterns, word lists and model of a de-identification system. The judge scores recall of identifiers on annotated notes they never see, with precision held.
For Enterprises
- Measures
- Recall of identifiers
- Holds
- Precision, name recall and speed held
- Hidden
- Notes from another department
Clinical evidence review
Agents read every registry record and publication on one PICO question and record claims with exact quotes. Code checks each quote and number, and a review panel of models decides each question.
For Enterprises · Solo builders
- Counts
- Verified claims, trial by trial
- Checked
- Each quote and number found in its archived record or paper
- Decided
- A panel of models, by majority
FHIR API conformance
Agents change the server's FHIR mapping, searches and export. The judge starts the server, runs Inferno's test groups and the HL7 validator, and the run ends when every goal passes twice.
For Startups · Enterprises
- Measures
- Inferno test groups passed
- Holds
- No validator errors, every value from its source record
- Hidden
- Patients from a seed the agents never see
Guides by audience
Twenty minutes to a first verdict.
Install it on one computer, or ask your own AI agent to:
Set up Thermite on this computer for me: follow https://usethermite.com/llm-setup.txt