Go from features and experiments to served, monitored models — with the engineering discipline production teams expect.
Most models never reach users because nobody versioned the data, tracked experiments or planned for drift. ML engineers turn promising models into services that keep working after week one.
Every week has a hands-on lab and an assignment. Open a week to see both.
Build models you can trust
Profile a card transactions dataset and document its quirks.
Write an EDA report with the distributions that matter for fraud.
Train a baseline and a gradient-boosted fraud classifier.
Compare models with cross-validation and justify your pick.
Segment merchants by behaviour with clustering.
Flag anomalous merchants and explain what makes them unusual.
Build velocity and time-window features for transactions.
Audit your pipeline for leakage and document the fixes.
Choose and evaluate the right model
Plot a cost curve for your fraud classifier.
Choose and document a decision threshold from the cost curve.
Train a PyTorch network on tabular transaction data.
Compare it against your gradient-boosted model.
Classify support tickets with pretrained embeddings.
Fine-tune a small model and report per-class errors.
Track every experiment in MLflow and register the best model.
Reproduce your best run from a single command.
Serve, monitor and improve
Serve the fraud model behind a scoring endpoint.
Add input validation, health checks and a load test.
Containerise the service and deploy it to the cloud.
Ship a CI pipeline that can roll back a bad model.
Detect drift on a shifted dataset with Evidently.
Write a retraining plan with clear triggers.
Agree scope and milestones for your capstone system.
Deliver the model card, runbook and demo.
The fast track covers the core in 12 weeks. The standard pace spreads the same content and the capstone over up to 16 weeks.
Plan for 8–10 hrs/week, including the live session.
A cost-aware fraud classifier with calibrated scores and a documented threshold.
Labelled card transactions with heavy class imbalance
A probability-of-default model with explanations suitable for a credit committee.
Loan applications and repayment history
{
"transaction_id": "txn_8816",
"score": 0.87,
"decision": "review",
"latency_ms": 41
}A containerised real-time scoring service with health checks and latency SLOs.
Live transaction payloads
Ship a deployed ML system that scores card transactions in real time, logs predictions, detects drift and documents how and when to retrain.
Dataset: Card authorisations with delayed fraud labels
Prefer another domain? Choose SaaS, E-commerce, Logistics or Healthcare at the capstone kick-off.
Build and deploy ML systems
Solve product problems with ML
Run training, serving and monitoring
These are the roles this track prepares you for. AI4Impact provides career support; it does not guarantee employment, interviews or a particular salary.
Completion records that you did the work. The professional certificate is awarded only after an independently reviewed capstone and a recorded defence, so it means something to the people hiring.
Add it to any track. Five modules that take you from a finished capstone to real interviews, for ₹24,999.
Each week has short recorded lessons, a hands-on lab and an assignment, plus a live clarification session with your trainer. Plan for 8–10 hours a week, including the live session.
They are part of Programme+: five one-to-one 30-minute sessions you book through your learner dashboard and use for scoping, reviews or career direction.
Fintech is the default. If you prefer another domain, you can choose SaaS, E-commerce, Logistics or Healthcare at the capstone kick-off.
No. AI4Impact provides career support and recommendations, but it does not guarantee employment, interviews or a particular salary. Hiring decisions are made by employers.
Book a free 20-minute call. We look at your background and tell you honestly which track fits, or if none of them do yet.