RAG · Retrieval-augmented generation

04 / Retrieve

From finding documents to answering with evidence. Connect retrieval, citations, evaluation and trust boundaries.

  1. Documents & versions
  2. Chunks & index
  3. Retrieval & reranking
  4. Evidence & answers
  5. Evaluation & abstention

Two offline practices are published: a BM25 retrieval baseline and citation checks for closed fields. They do not establish real-model answer quality or constitute a complete blog Q&A system.

Try it: choose evidence and adjust ranking →

START HERE

Series overview · suggested reading order

  1. Engineering cases

    Java LLM VI: validate retrieval evidence before judging RAG answers

    Implement BM25 and citation boundaries over ten owned synthetic documents and fourteen frozen queries, retaining misses and false hits.

    8 MINPublished
  2. Engineering cases

    Java LLM practice (7): a real citation can still support a wrong answer

    22 frozen configuration cases separate citation existence, revision scope, field support and completeness in a Java answer gate.

    9 MINPublished

PROJECTS

Related projects

  1. PROJECT / 002

    Java LLM Practice

    Reproducible Java conversation, retry and persistence boundaries, alongside attention, cache-cost and quantization-error experiments.

    Building
Related experiments
  1. LAB / 001

    Retrieval quality comparison

    Use fixed questions and documents; compare retrieval hits, citations and answer correctness with and without retrieval.

    Running

Paper reading guides

Learning path & planned topics

These are future topics, not completed experiments.

  1. Freeze the corpus, then compare chunking · PLANNED

    Build traceable documents and relevance labels, separating development and held-out questions.

  2. Choosing BM25, dense or hybrid retrieval · PLANNED

    Compare lexical matching, dense retrieval and fused ranking on identical questions.

  3. Reranking and assembling the context · PLANNED

    Select evidence with answer coverage and limited duplication under a fixed context budget.

  4. From retrieved evidence to answer quality and abstention · PLANNED

    Measure retrieval hits, claim support and abstention on missing evidence separately.

  5. Updates, permissions and retrieval injection · PLANNED

    Prevent withdrawn, stale or unauthorized documents from becoming current answer evidence.

  6. Integration: traceable blog Q&A · PLANNED

    Connect retrieval, evidence, answers and failure handling in a reproducible minimal application.

Continue along a learning path →

Try it · Retrieval and ranking

Expand to explore in your browser. Link to this exercise ↗