Author

aiterms

What is hybrid search (BM25 + vector)?

Hybrid search is a retrieval method that combines keyword search with vector similarity search so you can find documents that match both the exact words a user typed and the broader meaning of the query.

Why it matters

Pure keyword search is good when the exact term matters. Pure vector search is good when wording varies but meaning is similar. In practice, many real queries need both.

You reach for hybrid search when:

• users may search with exact product names, error codes, or technical terms
• the same idea can be expressed in many different ways
• you want better recall than keyword search alone, without giving up precision from exact matches

A common pattern is search over docs, help centers, knowledge bases, and enterprise content.

How it works

Hybrid search usually runs two retrieval methods in parallel:

1. BM25 scores documents by matching query words against document terms. It favors documents with the same keywords, adjusted for term frequency and document length.
2. Vector search embeds the query and documents into vectors and retrieves documents with similar meaning, even when the words differ.

The system then combines the results. The combination can happen in a few ways:

• merge two ranked lists and re-rank them
• normalize the scores and add them together
• use one retriever to generate candidates and the other to refine them

There is no single universal formula for hybrid search. The important idea is that it blends lexical matching and semantic matching. That makes it more robust than either method alone on messy, natural-language search.

Tiny concrete example

A user searches for:

“how do I reset MFA after phone loss”

• BM25 may strongly favor documents containing “reset,” “MFA,” and “phone loss.”
• Vector search may also retrieve docs about “recover authentication app” or “re-enroll two-factor authentication,” even if those exact words are missing.

A hybrid retriever can return both kinds of documents and rank the most useful ones near the top.

Common pitfalls / when NOT to use it

• Don’t use it when you only need exact lookup. If users are searching IDs, filenames, or codes, BM25 alone may be enough.
• Don’t assume “more signals” always means better. Poor score fusion can make results worse if one side dominates.
• Don’t skip evaluation. Hybrid search needs tuning against your own queries and relevance judgments.
• Don’t confuse retrieval with reranking. Hybrid search finds candidates; a separate reranker may still be useful afterward.
• Don’t overbuild for tiny corpora. On small datasets, a simpler search stack may be easier to operate and just as good.

Related terms

• What is retrieval-augmented generation (RAG)?
• What is a vector database?
• What is an embedding?
• What is semantic search?
• What is chunking (and semantic chunking)?
• What is a reranker / cross-encoder?
• What is top-k retrieval?
• What is a knowledge base for RAG?
• What is contextual / late chunking?
• What is RAG vs fine-tuning?

同じ著者の記事

What is chunking (and semantic chunking)?

Chunking is the practice of splitting a larger piece of text or data into smaller pieces so a system can process, store, or retrieve it more effectively; semantic chunking does the same thing, but tries to cut at meaningful boundaries instead of by fixed length. Chunking is a core technique in retrieval-augmented generation (RAG), document search, summarization, and any workflow that needs to feed long content into an LLM. Why teams use it: LLMs have context limits, so very large documents must

papoo.work

What is semantic search?

Semantic search is a way to find information by meaning, not just by matching the exact words you typed. Traditional keyword search is good when you know the exact term to look for. Semantic search helps when users phrase the same idea in different words, use synonyms, or ask a question instead of naming a document. You’d reach for it when: search queries are messy or conversational, exact wording varies a lot across documents, you want to retrieve “about the same thing” rather than “contains th

papoo.work

What is an embedding?

An embedding is a vector embedding: a list of numbers that represents something—like a word, sentence, image, or user—in a way a model can compare mathematically. Embeddings turn messy, human data into a form machines can use for: Similarity search: “Find documents like this one.” Recommendations: “Show items close to what the user liked.” Clustering and classification: “Group related things together.” Retrieval for LLMs: “Fetch the most relevant context before generating an

papoo.work

What is a vector database?

A vector database, also called a vector store, is a system for storing and searching data by its numerical embeddings so you can find items that are “most similar” rather than exactly matching text or IDs. Regular databases are great when you know the exact thing you want: a user ID, a product SKU, a timestamp range. But many AI features need semantic search or nearest-neighbor retrieval: “find questions like this one,” “retrieve passages relevant to this prompt,” or “recommend items

papoo.work

What is retrieval-augmented generation (RAG)?

Retrieval-augmented generation, or RAG, is a way to make a language model answer questions using relevant external documents it retrieves first, instead of relying only on what it memorized during training. RAG helps when you want answers that are: grounded in your own content, such as docs, policies, tickets, or product knowledge easier to update than retraining a model more likely to mention current or proprietary information In practice, teams reach for RAG when a model needs to answer “what

papoo.work

What is tool use in AI agents?

Tool use in AI agents is when an LLM-based agent can call external functions, APIs, or programs—often called a tool calling agent—to get information or take actions it cannot do reliably from text alone. A plain language model is good at generating text, but it does not reliably know live facts, cannot directly query a database, and should not guess at actions like sending email or creating tickets. Tool use solves that by letting the model delegate specific steps to purpose-built systems. Y

papoo.work

What is the agent loop (perceive-plan-act)?

The agent loop, also called the perceive-plan-act loop, is the repeated cycle where an AI agent observes its situation, decides what to do next, and then takes an action. This is the basic control pattern behind many agentic systems. Instead of asking a model for one one-off answer, you let it keep checking the world, updating its plan, and acting until the task is done or it should stop. You reach for it when the task is dynamic, multi-step, or depends on intermediate results: calling tools, br

papoo.work

What is a ReAct agent?

A ReAct agent is an LLM-driven system that alternates between reasoning and acting—often called reason + act or ReAct prompting—so it can decide what to do next, use tools, and update its plan from the results. ReAct is useful when a model should not just produce an answer, but also take steps in the world: search a document set, call an API, inspect a database, or ask for more information. In practice, teams reach for it when: a plain chat completion is too static, the task

papoo.work

What is an agentic workflow?

An agentic workflow is a workflow where an AI system can decide and take multi-step actions toward a goal, instead of only producing a one-shot answer; you may also hear this called an agentic system. Use an agentic workflow when the task is not just “generate text,” but “figure out what to do next.” That includes tasks like investigating an issue, pulling together information from multiple tools, retrying after failure, or following conditional branches based on intermediate results

papoo.work

What is an AI agent?

An AI agent, also called an LLM agent or autonomous agent, is a system that can pursue a goal by deciding what to do next, taking actions, and using the results to keep going. Most AI systems just answer a prompt once. An agent is useful when one answer is not enough and the system needs to act over multiple steps: search, call tools, inspect results, retry, and stop when the goal is met. You’d reach for an agent when the task is: open-ended, not a single fixed response dependent on external

papoo.work