Curated · Vector Databases
Top 50 Vector Databases Interview Questions
Here are the top 38 AI interview questions on Vector Databases, sorted by community quality. Each entry links to a detailed answer with explanations, hints, and source-grade follow-ups. See all in this topic →
Sorted by community quality · refreshed as new questions are published.
- 01HNSW vs IVF, when do you pick each for a production vector index?FlashcardMedium·Qual 4.5PineconeQdrant
- 02Match each vector index type to its primary characteristic.Match PairsMedium·Qual 4.5Pinecone
- 03Why is cosine similarity preferred over Euclidean distance for text embeddings?FlashcardEasy·Qual 4.4CohereOpenAI
- 04Pick the right vector database for a startup with 2M documents, no DevOps headcount, and a one week launch deadline.Multiple ChoiceEasy·Qual 4.0Pinecone
- 05Where does Hamming distance become the natural metric inside a vector database, and how is it implemented efficiently?FlashcardEasy·Qual 4.0Qdrant
- 06Why does every production retrieval system use approximate nearest-neighbor search instead of exact search past a certain corpus size?Multiple ChoiceEasy·Qual 4.0Pinecone
- 07Explain exactly what the ef_search knob on an HNSW index controls at query time.FlashcardEasy·Qual 4.0Pinecone
- 08Fill in the IVF nlist rule of thumb.Fill in BlankEasy·Qual 4.0Pinecone
- 09pgvector ships two index types. Which should be your default reach in 2026, and why was it added later than the other?FlashcardEasy·Qual 4.0
- 10Name two situations where Flat brute force search is still the correct production choice in a 2026 vector pipeline.Short AnswerEasy·Qual 4.0Pinecone
- 11Select all dimensions that meaningfully drive cost on a managed vector database bill in 2026.Multi-selectEasy·Qual 4.0Pinecone
- 12Cosine similarity and dot product give identical rankings under exactly one condition. Which is it?Multiple ChoiceEasy·Qual 4.0Pinecone
- 13At what scale does 'just put it in Postgres with pgvector' stop being the right answer?Multiple ChoiceMedium·Qual 4.0
- 14Walk through how Product Quantization compresses a 1024-dim float32 vector down to 32 bytes, step by step.Short AnswerMedium·Qual 4.0Zilliz
- 15Modern vector databases combine sparse (BM25) and dense vectors at the storage layer via two distinct patterns. Identify them.Multiple ChoiceMedium·Qual 4.0Pinecone
- 16Sharding a vector index that does not fit on one node, how is it typically done, and what's the query-time cost?Short AnswerMedium·Qual 4.0PineconeQdrant
- 17What does Matryoshka representation learning buy a vector-database operator beyond regular embeddings?Multiple ChoiceMedium·Qual 4.0Pinecone
- 18As M grows in an HNSW index, four things move at once. Identify what climbs versus what slows.Multiple ChoiceMedium·Qual 4.0Pinecone
- 19Spot the conceptual error: 'We're using PQ instead of HNSW for our vector index because PQ is faster.'Spot the ErrorMedium·Qual 4.0Zilliz
- 20Pre-filter versus post-filter strategies for metadata-constrained ANN search — what breaks at each extreme?Multiple ChoiceMedium·Qual 4.0PineconeQdrant
- 21PQ-based distance computation outruns full float32 distance even after decoding the vectors. Why?Multiple ChoiceMedium·Qual 4.0Zilliz
- 22Spot the error in this candidate's description of IVF's two main parameters.Spot the ErrorMedium·Qual 4.0Pinecone
- 23Predict the shape of recall and latency on an IVF index of 10M vectors as nprobe sweeps from 1 to 50.Predict OutputMedium·Qual 4.0Pinecone
- 24When does binary quantization beat Product Quantization as the production compression choice for a vector index?Multiple ChoiceMedium·Qual 4.0Qdrant
- 25Doubling the embedding dimension from 768 to 1536 — what changes in storage, query latency, and PQ trainability?Multiple ChoiceMedium·Qual 4.0Pinecone
- 26Define recall@10 for a vector index, and spell out exactly what 'ground truth' means in that definition.Short AnswerMedium·Qual 4.0Pinecone
- 27A team indexed embeddings under cosine but queried with Euclidean, and recall collapsed. Diagnose the cause.Short AnswerMedium·Qual 4.0Pinecone
- 28How does an HNSW index handle deletes and updates under a heavy write workload?FlashcardMedium·Qual 4.0Qdrant
- 29Explain what NVIDIA CAGRA buys you over CPU HNSW and the operational catch that decides whether you should reach for it.FlashcardMedium·Qual 4.0Zilliz
- 30Match each vector-compression scheme to its core mechanism.Match PairsMedium·Qual 4.0Qdrant
- 31Which of these 2026 vector databases are credible at billion-vector scale? Select all that apply.Multi-selectMedium·Qual 4.0PineconeQdrant
- 32Estimate the RAM cost of an HNSW index over 100M vectors at 1024 dimensions with M=32, including graph overhead.Short AnswerHard·Qual 4.0Pinecone
- 33Compare namespace-per-tenant against single-shared-index-with-tenant-filter for a B2B RAG product serving 5000 customers.Short AnswerHard·Qual 4.0Pinecone
- 34A vendor's benchmark chart plots QPS against recall@10. What should you look for, and what claim should make you skeptical?Multiple ChoiceHard·Qual 4.0Pinecone
- 35Aggressive metadata filtering can silently destroy recall on an HNSW index even when matches exist. Why?Short AnswerHard·Qual 4.0Qdrant
- 36What problem does DiskANN solve that HNSW does not, and when do you reach for it?Multiple ChoiceHard·Qual 4.0PineconeZilliz
- 37Identify the workload pattern where a traditional ANN-indexed vector database is the wrong tool, and a log-structured or batch-rebuild approach fits better.Multiple ChoiceHard·Qual 4.0Qdrant
- 38A team's Pinecone bill 5x'd month over month with no traffic change. Order the most likely causes to investigate, from most likely to least likely.Order StepsHard·Qual 4.0Pinecone
38 of 50 requested: only 38 published in this topic so far.
Related top picks
By experience level
Other topics
Top 50 Inference Optimization Interview Questions→Top 50 Fine-Tuning Interview Questions→Top 50 LLM Evaluation Interview Questions→
By role