Lunai treats machine learning as software: versioned data, versioned code, versioned models. Notebooks are a workshop, not the deliverable.
ML that operations will touch
Scoring APIs, batch scores into the warehouse, or a simple screen for the planner. If the user cannot explain the top reasons, we keep the model simpler.
- Classification, ranking and forecasting with monitoring
- Feature pipelines that match the training grain
- Canary releases and quiet hours for first production scores
- Retrain triggers you agree in writing
Consulting, then build
Some problems need a rule, not a model. We will say so. When ML is right, we staff data + engineering together so the job is not stranded on a data-science island.
- Problem framing with the decision owner
- Label strategy and bias checks
- Security review for training data
Machine learning from Lunai is for teams in New Delhi and across India who need a number in the system of record, not a conference talk.