Applied machine learning
Production ML for healthcare, marketplaces, and behavioral data.
Intovoid AI builds machine learning systems end to end — from data pipelines and model training through to deployment in the software people already use.
Focus
MKB-10 coding from Serbian clinical notes
Our first healthcare product reads a physician-written note and lab values, then suggests the three most likely MKB-10 diagnosis codes with confidence scores. Built in partnership with the Cigota Institute.
- Training protocols
- 105,000
- Top-3 accuracy
- 82%
Accuracy measured on held-out test data.
Clinical note
Pacijent, 58 god. Žali se na pojačanu žeđ i zamor unazad dva meseca. Glukoza našte 9.8 mmol/L, HbA1c 7.9%.
- E11.9Dijabetes melitus tip 2, bez komplikacija0.87
- R73.0Abnormalni test tolerancije glukoze0.08
- E11.6Dijabetes tip 2 sa drugim komplikacijama0.03
What we work on
Three domains, one way of building
- 01
Healthcare
Clinical NLP for Serbian medical records. Diagnosis coding, structured extraction, and models that tolerate OCR noise from scanned paper.
- 02
Marketplaces
Recommendation systems and price prediction models that run against live inventory and real user traffic.
- 03
Behavioral data
Modeling how people actually use consumer platforms, turning event streams into features that drive decisions.
How we build
From raw data to running in production
- 01
Data
Ingestion, cleaning, and labeling pipelines built for messy real-world sources.
- 02
Training
Model selection, training, and evaluation on held-out data that reflects production.
- 03
Deployment
Local or cloud inference, integrated into the systems teams already use.
Contact
nikola@intovoid.dev- PIB
- 114731407
- MB
- 67811305