ImprovementsSince moving to DagsHub, Beewise has seen major improvements:Model iteration time dropped from 2–3 weeks to just 3–5 days.Debugging time went from days to just a few hours.Annotation mistakes decreased by ~30%.Active learning management: 1 annotator now handles 100,000+ images
Improvements
Since moving to DagsHub, Beewise has seen major improvements:
Model iteration time dropped from 2–3 weeks to just 3–5 days.Debugging time went from days to just a few hours.Annotation mistakes decreased by ~30%.Active learning management: 1 annotator now handles 100,000+ images independently.
In Boaz, AI Lead, had this to say:
“DagsHub turned our chaotic data management into a structured, reliable system — we move 3× faster and fix issues in hours, not days.”
By cleaning up data workflows and giving the team full visibility into experiments, Beewise was able to dramatically speed up model development and focus on improving beehive health at scale.
About Beewise
Beewise is on a mission to protect global bee populations using robotics, AI, and computer vision. Their robotic beehives monitor hive health, detect problems early, and even take action automatically — allowing for 24/7 hive care at scale without human intervention.
Data Types
Beewise mainly works with vision-based data:
High-resolution images of hive interiors (combs, bees, brood)Robotic control vision data for precise mechanical interactionsExpert-labeled annotations, often thousands per image, to capture subtle biological signals
These large datasets form the foundation for training models to both monitor and physically interact with beehives.
Challenges
As Beewise scaled, they initially relied on Airtable to manage experiments and data. This setup quickly became a bottleneck:
No version control: They couldn’t trace how data changes impacted model performance.Annotation mistakes: Manual errors across thousands of labels made models unreliable.Slow debugging: Diagnosing model failures could take days to weeks.Growing complexity: Each new model and dataset increased the risk of copy-paste errors and data drift.
As the number of tasks and amount of data grew, these issues made it harder to maintain model quality and slowed down their AI development cycle.
Solutions
Beewise moved their data and experiment management to DagsHub to solve these challenges. Using DagsHub, Beewise was able to:
Automatic versioning captured every change to datasets and experiments.Structured workflows made it easy for annotators to filter, sort, and manage large datasets.Active learning loops became smooth:The model flags errors → annotators correct → retrain quickly → repeat.Instant rollback allowed the team to compare dataset versions and debug regressions easily.
This shift brought structure, traceability, and speed to their AI pipeline.
