Computer Vision Interview Questions and Answers 2026 (55 Questions)
55 computer vision interview questions answered: CNNs, ViTs, detection, mAP, NMS, segmentation, SAM, CLIP, OCR, edge quantisation, DPDP privacy and scenarios.
Learn
Guides, career roadmaps and explainers from Cloud Soft Solutions trainers. 667 articles.
55 computer vision interview questions answered: CNNs, ViTs, detection, mAP, NMS, segmentation, SAM, CLIP, OCR, edge quantisation, DPDP privacy and scenarios.
65 deep learning interview questions with answers: backprop, AdamW, batch vs layer norm, CNNs, RNNs, transformers, PyTorch, mixed precision, DDP and scenarios.
60 Python interview questions for AI engineers: async and semaphores, Pydantic v2, FastAPI streaming, pytest mocks, the GIL and six live-coding prompts.
Build a procurement AI agent for PO exceptions: supplier emails, read-only ERP data, contract checks, buyer-approved replies, spend limits and replay evals.
Build a claims triage AI agent for motor and health FNOL: extraction, policy checks, missing documents, fraud signals, routing, DPDP, fairness and evaluation.
AI content provenance explained: C2PA manifests, Content Credentials, SynthID-style watermarks, why detection fails, EU AI Act and India rules, what to build.
Codebase RAG explained: syntax-aware chunking, hybrid code search, code graphs, agentic search vs embeddings, index freshness, repo permissions and evaluation.
Build an AI meeting assistant: evidence-backed decisions and action items, owner-confirmed Jira tasks, cited answers, consent, opt-outs, access and retention.
How to build durable AI agents: event history and replay, LLM calls as activities, Temporal, Step Functions, LangGraph durability modes, versioning and cost.
Build call center QA AI end to end: code-mixed transcription, PII redaction, evidence-cited scorecards, compliance flags, human review, fairness checks and ROI.
AI incident response playbook: incident types, severity levels, kill switches, rollback, trace-based investigation, notification duties and tabletop exercises.
What an AI product manager really does: evals as the spec, the AI PRD, launch criteria, cost per interaction, metrics, mistakes and a 90-day plan to move in.
Enterprise knowledge management for AI: content ownership, review cycles, authoritative sources, writing for retrieval, metadata, conflicts and a RAG checklist.
GPU for AI engineers explained: VRAM requirements for LLMs, memory bandwidth, FP8/BF16, NVLink, quantisation, multi-GPU, batching, cloud options and cost.
Model distillation explained: soft-label vs LLM distillation, when it pays off, the teacher-student workflow, data pitfalls, provider terms and managed options.
Agentic commerce explained: how AI agents buy and pay safely with AP2 mandates, ACP, Visa and Mastercard agent programmes, UPI Reserve Pay pilots and limits.
On-device AI explained for enterprise: why run models at the edge, limits, runtimes, quantisation, hybrid cloud fallback, fleet updates, security and OT.
How computer use agents and browser agents work, when to use them instead of APIs or MCP, how they relate to RPA, and the controls that make them safe enough.
Open-weight LLMs for enterprise: open-weight vs open-source AI, licence types, acceptable use, fine-tune obligations, supply-chain risk and an approval process.
AI agent protocols explained: where MCP, A2A, AG-UI and payments protocols like AP2, ACP, UCP and x402 sit, who stewards them, maturity and what to learn first.
50 LLM evaluation interview questions with answers: golden sets, RAG metrics, LLM-as-a-judge, agent evaluation, CI regression gates and real scenarios.
LangChain 1.0 migration guide: move AgentExecutor to create_agent, memory to checkpointers, legacy chains off langchain-classic, community to partner packages.
50 AI system design interview questions with answers: requirements, token and cost estimates, RAG, agents, LLM gateway, text-to-SQL, voice bots and evaluation.
60 AI security interview questions with answers: OWASP LLM Top 10, prompt injection, agent identity, RAG ACLs, guardrails, red teaming and 12 real scenarios.