bKash bTechWhiz
Advanced Research Engineer viva — pipeline & question archetypes
The second round is a viva voce; candidates selected from this round advance to the semi-final, also held on the 14th. The panel runs as a standard viva board — ML-based, small code possible, conceptual questions. Each archetype below has matching drill questions in its linked module.
Six question archetypes
Conceptual rapid-fire
Bias/variance, overfitting, regularization, precision/recall, cross-validation, supervised vs unsupervised — answered out loud in 60-90 seconds.
Math for ML
Bayes theorem word problems, eigenvalue/rank intuition, expectation/variance, distributions, derivative of sigmoid → chain rule → backprop.
Small coding
On paper, no IDE: reverse linked list, valid parentheses, two sum, plus ML-flavored numpy-from-scratch (KNN, train/test split, accuracy).
Scenario & system design
CCTV car counting, classifying unlabeled articles at scale, diagnosing loss curves, fraud detection for a fintech wallet.
Research interpretation
Summarize a paper (your thesis or a real ML paper): problem → method → findings → real-world implication, in under 2 minutes.
Fintech domain
What ML problems does an MFS company have? Fraud/anomaly detection, credit scoring, churn, recommendation, OCR/KYC, chatbots.
Your actual experience (fill in after 14 July)
Come back after the viva and log what was actually asked, which prep questions matched, and what you'd do differently for future candidates — this is what turns this into Interview-BD-style crowd-sourced value.