llms.txt
Compare with other sites' llms.txt files →# Krv Labs
> Krv Labs (Krv Labs, Inc.) builds tunable, private clinical trial intelligence: models that learn what a plausible patient looks like for a specific trial, adapt to each site's longitudinal records, and run locally in the hospital's own environment with zero data transfer.
## Core Platform: Tunable, Private Clinical Trial Intelligence
- [Clinical Trial Intelligence (Home)](https://krv.ai/): Protocol-tuned models that surface candidates static filters miss across longitudinal EHR records, clinical notes, labs, and trajectories.
- [Erdos](https://erdos.krv.ai): Clinical trial intelligence and cohort discovery platform.
- **How it works**:
1. **Learn the patient landscape**: Unsupervised discovery on broad, anonymous data reveals high-value patient cohorts across care, documentation, and biology simultaneously, validated with built-in clinician review.
2. **Algorithmic protocol matching**: Learned cohorts are scored against trial criteria with inspectable rationale for every match.
3. **Read trajectories, not snapshots**: Time-series foundation models read longitudinal lab and clinical trajectories to separate true candidates from likely screen failures.
4. **Fine-tune custom models**: Protocol-specific model packs are fine-tuned on open-source clinical data to capture complex signals across notes, labs, and trajectories.
5. **Screen in a desktop app (100% local & private)**: Hospitals drop a standard EHR export into a local desktop app on their own machine. Fine-tuned model packs rank candidates locally with zero PHI transfer and no EHR integration required.
6. **Investigator review**: Investigators receive a prioritized shortlist mapped to protocol criteria with forecasted eligibility and inspectable evidence.
- **Deployment speed**: 1–3 weeks in parallel across sites (skipping or reducing multi-month data access, BAA, security audit, and HL7/FHIR EHR integration bottlenecks because no PHI leaves the site).
- **Evidence**: In a diagnosis-blinded evaluation across 9,172 records, Krv identified 91.6% (92/100) of known target cases vs. 56.5% with standard code-based filters.
## Open-Source Stack
- [Open Source Hub](https://krv.ai/open-source): Scientific and clinical AI software built in the open by Krv Labs.
- [Topos](https://krv.ai/topos): Structural code quality and security evaluation engine for agent-generated code (`curl -fsSL https://krv.ai/topos/install.sh | sh`).
- [Pulsar](https://krv.ai/pulsar): Multi-view topological structure discovery (Thema algorithm, published in *Nature Energy* 2025) implemented in Rust and Python (`uvx --from thema-pulsar pulsar install`).
- [Phil (`philler`)](https://docs.krv.ai/phil): Topological multiverse imputation for clinical and tabular EHR data (`uv pip install philler`).
- [TRAILED](https://docs.krv.ai/trailed): Differentiable Euler Characteristic Transforms (ECT) for longitudinal and tabular EHR data (`uv pip install trailed`).
- [Pasteur](https://krv.ai/pasteur): Open-source clinical AI stress-testing toolkit in Rust and Python (`uv pip install pypasteur`) simulating Blackout, Jitter, and Flipper perturbations.
- [Phirewall](https://krv.ai/open-source#healthcare-safety): Local PHI vault and fail-closed MCP gateway using deterministic Tokenized Twins so AI assistants can analyze clinical data without seeing patient identities.
- [Ethos](https://krv.ai/open-source#healthcare-safety) (coming soon): MCP server for ethical study planning. It structures study design with a researcher before any analysis and never touches patient data.
## Company & Research
- [Team](https://krv.ai/team): Founded by Sidney Gathrid (CEO), Jeremy Wayland, PhD (CSO), and Stuart Wayland (CTO). Backed and supported by NVIDIA Inception, Berkeley SkyDeck (PAD-13), and TUM Venture Labs Healthcare.
- [Field Notes](https://krv.ai/field-notes): Technical research, clinical AI evaluations, and peer-reviewed publications.
- [News](https://krv.ai/news): Company milestones, product releases, and research announcements.
- [Documentation](https://docs.krv.ai): Technical documentation for Krv's open-source tools.
- [Book a Protocol Review / Demo](https://kalendr.krv.ai/book/c-ayk3u): Schedule a clinical trial protocol feasibility review with Krv Labs.