AI Engineering Interview Questions — Practice Library | Velocode

300+ problems

Problem

Tags

Type

Stream Aggregator Batches LLM Tokens Into Complete Sentences

Reported at GoogleSWE (AI-Era)Medium

Conversation History Compression for Long-Running Agents

Reported at AnthropicSWE (AI-Era)Hard

Build a Connector That Respects a Vendor Rate Limit

Forward DeployedMedium

Hierarchical cache breakpoints: system prompt, tool definitions, context windows, query batches

Reported at CursorAI EngineeringMedium

Template variable substitution with validation and defaults

Reported at CohereSWE (AI-Era)Easy

Build a Health Check So the Customer Can Trust the Deployment

Forward DeployedMedium

Retrieval deduplication: chunks from same document not filtered before ranking.

Reported at PineconeAI EngineeringMedium

Code Comprehension

RAG Chatbot Ignoring Retrieved Documents, Using Training Data Instead

Reported at AnthropicAI EngineeringEasy

Code Comprehension

A Requested Feature Conflicts With Data Privacy

Forward DeployedHard

Conversation history manager fails to truncate old messages properly

Reported at CohereSWE (AI-Era)Medium

Code Comprehension

Hallucination Detection Pipeline Fails On Contradictory Claims Against Context

Reported at Scale AIAI EngineeringMedium

Code Comprehension

Agent Calls the Wrong Tool

Forward DeployedMedium

Code Comprehension

Access control filter placement before reranker causes leakage bug

Reported at CohereAI EngineeringMedium

Code Comprehension

Evaluation Metric Misalignment: Faithfulness Scores Mask Semantic Errors

Reported at AnthropicAI EngineeringMedium

Code Comprehension

Re-Architect Retrieval to Scale to 2 Million Documents

Forward DeployedHard

Fine-tuned model tokenizer mismatch with deployment environment configuration

Reported at MetaAI EngineeringHard

Code Comprehension

Token Budget Manager for Multi-turn LLM Applications

Reported at OpenAISWE (AI-Era)Easy

Build a Tool to Triage Incoming Customer Feedback

Forward DeployedMedium

Vector Embedding Normalization Missing Causes Low Similarity Scores

Reported at OpenAIAI EngineeringEasy

Code Comprehension

LLM Agent Tool Calls Exceed Context Limit Token Budget

Reported at Google DeepMindAI EngineeringMedium

Code Comprehension

Stop the Agent Inventing Facts in a Regulated Domain

Forward DeployedHard

Code Comprehension

Sliding Window Rate Limiter for Concurrent LLM API Calls

Reported at DatabricksSWE (AI-Era)Medium

Prompt Injection Vulnerability in Content Filter Implementation

Reported at AnthropicSWE (AI-Era)Medium

Code Comprehension

Build a Robust CSV Importer for Messy Customer Data

Forward DeployedEasy

Loading more problems…