SCIENCE
The science behind XERJ
These pages explain the technical terms XERJ uses in its products, in plain language. Each entry is a short definition, then how the idea works, how XERJ uses it, and the research it rests on. When a how-to already lives under Answers, that page stays the practical guide; the science page is the concept.
Retrieval
What is hybrid search?Hybrid search runs a full-text query, such as BM25, and a vector query on the same question, then merges the two ranked lists into one result list.
Hybrid vs vector vs semantic searchVector and semantic search rank by meaning, usually as vectors. Hybrid search also runs a full-text arm, such as BM25, and fuses the two ranked lists.
Reciprocal rank fusion formulaThe reciprocal rank fusion formula is score = 1 / (k + rank), added across every list an item appears in. The lists need no shared score scale.
BM25 vs vector searchBM25 vs vector search: BM25 ranks by query terms, how rare they are, and document length. A vector query ranks by meaning, so a paraphrase can still match.
What is semantic search?Semantic search ranks passages by meaning, usually as vectors, so a paraphrase can match even when it shares few words with the query.
What is reranking?Reranking is a second stage after retrieval: a slower model reorders a short candidate list, and does not walk the whole index again.
Models and checking
Neurosymbolic AI vs LLMsA language model proposes an answer in its own text. Neurosymbolic AI also runs a program or formula and checks the result against facts.
What are neurosymbolic models?A neurosymbolic model pairs a neural part that proposes a program or formula with a symbolic part that runs it and checks the result against facts.
What are grounded answers?A grounded answer is one where every claim traces back to a source you can open and check. How grounding works, how to check it, and how XERJ does it.
What is extractive answering?Extractive answering takes the answer from the words of the document, as a quoted span or a value computed from spans, instead of writing new text.
Agents
Reference coding for AI coding agentsIn AI coding, reference coding is how an agent looks up an API it does not already know: it indexes libraries and reads a cited passage before it writes.
What is context engineering?Context engineering is choosing what an agent actually reads: a short, cited passage instead of a whole file, so the context window is spent on the task.
What is agentic memory?Agentic memory is memory an agent writes and reads itself through tools, so a fact can survive past one chat, not only through automatic hooks.
What is the Model Context Protocol?The Model Context Protocol, or MCP, is an open protocol that lets a host such as Claude Code or Cursor call tools on a node, including XERJ search.
Code and corpora
What is syntax-aware search?Syntax-aware search indexes declared symbols from a parser, so you can look up a definition by kind, name and line, not only by matching raw text.
What is Corpus Hub?Corpus Hub is XERJ's registry of signed, licence-reviewed index packs. An agent can ask a public corpus with file:line citations, without cloning the tree.