GenAI and LLM interview questions

Self-attention uses queries, keys, and values from the same sequence.

Every token attends to every other token, so scores scale as n times n.

During decoding, keys and values for past tokens are reused instead of recomputed.

BERT sees tokens on both sides of a masked position, which helps disambiguation for NLU.

Masked LM hides random tokens and reconstructs them from the full sentence.

State the role, allowed tools, output schema, and what to do when information is missing.

Prompts are code.

ReAct interleaves reasoning traces with actions such as search or calculator calls.

You constrain the model to emit valid JSON or schema-shaped text.

The query may be rewritten, then hybrid search returns top chunks, optionally reranked.

Query rewriting uses an LLM or rules to expand, correct, or split the user question into better search strings.

A reranker scores the first-stage hits more carefully, often with a cross-encoder.

HyDE generates a hypothetical answer, embeds that, and searches with the hypothetical embedding.

You generate several paraphrased queries, retrieve for each, and merge unique chunks.

Retrieval can miss the right chunk, return conflicting sources, or pack so much text that the model ignores the key sentence.

Re-embed changed chunks, upsert by stable IDs, and delete vectors for removed content.

You restrict candidates by tenant, date, product, or ACL before or during ANN search.

HNSW is a graph index with strong recall and simple incremental inserts, at higher memory cost.

Cosine cares about angle, L2 about Euclidean distance, inner product about both direction and magnitude.

Providing relevant passages and requiring citations reduces unsupported claims.