NaviMed-UMB
Clinical LLMs on hardware you own — local inference of open-weight models on consumer AMD R9700 GPUs, public model cards and an open benchmark. Data never leaves the building.
Open dashboard →Research & Lab
Projects at the intersection of pulmonology, functional diagnostics and artificial intelligence.
// log
- 2026-06 NaviMed-UMB: added a single-stream (N=1) anchor to the benchmark — the full “inference envelope” (single-stream + concurrency + thermal + power). dashboard →
- 2026-06 Scriptorium v1.3.0 — stat_run engine (permutation tests, multiple-comparison control). Public. GitHub →
- 2026-06 NaviMed: English SmPC (EMA) calibration corpus — open dataset + Zenodo DOI.
- 2026-05 PLLuM-70B: public family of 8 AWQ (W4A16) quantizations with model cards — first vLLM-native AWQ of the whole family. Hugging Face →
- 2026-04 NaviMed-UMB: hardware-envelope benchmark v0.1.0 — open, Zenodo DOI. Zenodo →
// projects
Ego Architecture AI
An interdisciplinary concept linking neuroscience, psychoanalytic theory and AI system architecture (draft v0.2).
Open dashboard →Capno-Nome
Artificial intelligence in capnography — analysis of the tcPCO₂ signal and the capnogram (Pattern Burden, time-series). Work in progress (v0.3).
Broncho-Nome
A bronchoscopy nomenclature lexicon plus support for navigation and biopsy of lung lesions — thoracic oncology diagnostics. In preparation.
Scriptorium
A public Claude Code plugin (agents + skills) for the scientific-writing and peer-review workflow — deterministic, rigorous, offline; it never fabricates a number. Released publicly as v1.0.0 (MIT).
Code on GitHub →NIV → Lung Transplant
Non-invasive ventilation as a bridge to lung transplantation — a systematic review and an international survey study.
DIVE-PL
An AI tool-chain for building survey studies — reproducible, human-led + AI-assisted: the investigator designs and orchestrates while a chain (Console, GPT-5.5, RSDCap, Python/React dashboards) executes and reviews under human oversight. Open-science.
Open →