We Build the Tool and the Network for the Health Economy
Any health signal in. 128 numbers out. Your model on the network.
Your research becomes a product.
Univault Technologies LLC (Salt Lake City, Utah) builds the General Learning Encoder (GLE), a subject-invariant biosignal AI foundation model that achieved over 13x more improvement than the next best team at the NeurIPS 2025 EEG Foundation Model Challenge across over a thousand competing teams. GLE encodes any health signal — EEG, breathing, cardiac, voice, molecular sensors — into 128 DCT-II frequency coefficients. The company also operates ParagonDAO, the independent verification network for health models.
Two Products. One Mission.
The tool without governance is dangerous — unvalidated health models loose in the world. The governance without the tool is just a committee with nothing to govern. We built both.
BAGLE
The Lab Equipment
The universal encoder. Send any health signal — breathing, heart rate, voice, EEG, biosensor output — and get back 128 numbers. Train a classifier on those numbers. That's your model.
- 6 research models published
- 29% less error than competition benchmarks
- API opens April 2026 — build Model #7
ParagonDAO
The Peer Review Board
The governance network. Validates models before patients rely on them. Certifies builders. Ensures quality across every health application on the network.
- Published whitepaper: The Health Economy
- HF-Auth continuous security layer
- 10% network fee funds the mission — preventing loss of life
Models Built on GLE
Each model was built the same way: health signal in, 128 coefficients out, classifier trained. These are the first. Yours is next.
1,000+ samples validated
EEG Consciousness
Real-time consciousness state classification from brain signals
300 samples validated
Type 2 Diabetes
Metabolomics-based screening using serum biomarkers
1,751 samples validated
Parkinson's & Alzheimer's
Saliva-based Raman spectroscopy for neurodegenerative screening
4,200+ samples validated
COVID-19 Detection
Real-time Raman-based detection from saliva. No reagents required
2,693 samples validated
Breathing Patterns
Audio-based breathing pattern classification with <0.5s latency
From Researcher to Founder
You bring the biology and the data. We bring the math. You keep 90% of every classification.
Collect Signals
Breathing recordings, biosensor readings, voice samples, molecular data — whatever you study.
Encode
Send signals to BAGLE API. Get back 128 numbers per signal. The hard math is done.
Train & Validate
Train a classifier on those 128 numbers. ParagonDAO validates accuracy before patients rely on it.
Ship & Earn
Your model becomes a screening tool anyone can use. A patient breathes into their phone — your model answers.
Research Behind the Encoder
The General Learning Encoder (GLE) is a foundation model for frequency-domain health intelligence
Breathing as Biometric Identity
96.8% identification accuracy across 97 participants using nasal airflow patterns alone. The same encoder that classifies disease also verifies identity — zero additional power.
GLE: Universal Health Signal Encoder
29% less error than competition-winning solutions on subject-invariant health prediction. Works on new users immediately — no calibration, no retraining.
The Mission
10% of all network fees fund one thing: preventing loss of life. 988 crisis detection, community health screening, free GLE access for crisis organizations. The tool without the mission is just technology. The mission without the tool is just hope.
Latest Update
Subject Invariance Demonstrated. The Verification Network Is Live.
The hardest problem in biosignal AI — making models that work on people they have never seen — has a working solution. Our GLE encoder improved over 13x more than the next best team among over a thousand competitors at the NeurIPS 2025 EEG Foundation Model Challenge. Today we launch the ParagonDAO Verification Network.
Read the full post