CV Interview Preparation

MAANG-level প্রশ্ন।

🎬 গল্প দিয়ে শুরু

Google, Meta, NVIDIA, Tesla — CV engineer-এর interview ৪–৬ round। DSA, ML theory, CV-specific, system design, behavioral। এক জায়গায় সব preparation।

Round structure (typical MAANG)

Round 1 — Coding (LeetCode medium/hard, 2 প্রশ্ন/45 min)।
Round 2 — ML fundamentals (bias-variance, overfitting, gradient)।
Round 3 — CV depth (CNN math, YOLO loss, ViT architecture)।
Round 4 — ML system design (build YouTube recommendation/CV pipeline)।
Round 5 — Behavioral (STAR format, leadership principle)।

DSA must-know (CV role-specific)

Arrays / strings (image as 2D array)।
Sliding window, two pointer।
BFS/DFS — connected components, flood fill।
Heap — top-K (NMS-এ ব্যবহার)।
DP basics, graph traversal।

Volume

১৫০ LeetCode Medium + ৫০ Hard = job-ready। NeetCode 150 শুরু করুন।

Core ML প্রশ্ন — সবসময় আসে

Bias vs variance, কীভাবে detect ও কমান?
Precision/Recall/F1 — কখন কোনটা priority?
Overfitting — ১০টি কৌশল বলুন।
Batch Norm internals — train vs eval mode পার্থক্য।
Dropout কীভাবে কাজ করে? Inference-এ scaling কেন?
Cross-entropy vs MSE — কেন classification-এ CE?
Adam vs SGD — কখন কোনটা?
Learning rate schedule — warmup, cosine, plateau।

CV-specific প্রশ্ন

Convolution-এ output size formula কী? Stride, padding দিয়ে calculate।
Receptive field কীভাবে বাড়ে — dilated conv-এর ভূমিকা?
YOLO loss function-এ ৩টি term কী? Anchor box কেন?
NMS algorithm step-by-step (soft-NMS ও বলুন)।
IoU vs GIoU vs DIoU vs CIoU পার্থক্য।
Semantic vs Instance vs Panoptic segmentation।
Faster R-CNN-এর RPN কীভাবে কাজ করে?
Vision Transformer-এ positional embedding কেন দরকার?
Data augmentation — Mixup, CutMix, Mosaic-এর সুবিধা।
mAP @ 0.5 vs mAP @ 0.5:0.95 কী?

ML System Design — কাঠামো

text

1. Requirements clarification (10 min)
   - Scale (QPS, dataset size, latency SLA)
   - Real-time vs batch
   - Accuracy vs cost trade-off

2. Data pipeline
   - Collection, labelling, augmentation, versioning

3. Model
   - Baseline → iteration → production architecture

4. Training infra
   - Distributed training, hyperparameter search

5. Serving
   - Model server (Triton), autoscaling, A/B test

6. Monitoring
   - Drift detection, retraining trigger, metric dashboard

7. Trade-offs ও alternatives discussion

Coding example — NMS

python

def nms(boxes, scores, iou_thr=0.5):
    idx = scores.argsort()[::-1].tolist()
    keep = []
    while idx:
        i = idx.pop(0); keep.append(i)
        idx = [j for j in idx if iou(boxes[i], boxes[j]) < iou_thr]
    return keep

def iou(a, b):
    x1, y1 = max(a[0], b[0]), max(a[1], b[1])
    x2, y2 = min(a[2], b[2]), min(a[3], b[3])
    inter  = max(0, x2-x1) * max(0, y2-y1)
    area_a = (a[2]-a[0]) * (a[3]-a[1])
    area_b = (b[2]-b[0]) * (b[3]-b[1])
    return inter / (area_a + area_b - inter + 1e-9)

Behavioral (STAR)

S — Situation (context)।
T — Task (responsibility)।
A — Action (specifically YOU did কী)।
R — Result (metric, impact)।
Top stories: conflict, failure, leadership, ambiguous problem।

৮ সপ্তাহের prep plan

text

Week 1-3: LeetCode 100 problem (Easy → Medium)
Week 4-5: ML/CV theory (Stanford CS231n + Bishop)
Week 6:   System design (5 mock design)
Week 7:   Mock interview (Pramp, interviewing.io)
Week 8:   Behavioral story + company-specific research

প্র্যাকটিস টাস্ক

NMS, IoU, depthwise conv নিজে scratch থেকে লিখুন।
"Design Instagram-এর AI photo enhancement" system design লিখুন।
Pramp-এ ৫টি mock interview করুন।