Research questions for practical business AI.
Anova AI Labs is a research track under AnovaGrowth for how smaller models, scoped agents and deployment choices should be evaluated before anyone recommends them. It documents questions and methods, not finished models or client outcomes.
Smaller models
Which compact model can be tested against a defined task, cost boundary, and acceptable failure mode?
Agent workflows
Which permissions, source context, and approval steps are required before a tool-using agent can act?
Self-hosted inference
When do local machines, workstations, or private environments provide a useful operating tradeoff?
Evaluation loops
What evidence, baselines, and review gates are needed before a model is recommended?
The research track records the task, evaluation conditions, operating limits, and review boundary a model would need before it can inform a client recommendation.
Task fit
A recommendation starts with a bounded task, not an abstract model comparison.
Operating limits
Privacy, cost, uptime, and hardware conditions stay visible in the method.
Review boundary
Sources, actions, and exceptions remain visible to the person responsible.
Clear labels
Research concepts stay separate from validated client evidence.
Related paths
Bring a bounded task and the constraints around it. We will say whether it is a research question or a build.