Design answers that can show their evidence
Choose an answer method, approve source versions, then test chunking, retrieval, access, index updates, citations, freshness, and no-answer behavior.
BEFORE LESSON ONE
Bring real material so the course produces a real result.
- 01 · WHO IT IS FOR
- For knowledge assistants, policy search, support systems, and internal research workflows.
- 02 · WHAT TO BRING
- Bring a small source set and five questions, including one with no eligible answer.
- 03 · HOW TO FINISH
- A reviewer can trace every important claim to current, permitted evidence without exposing restricted source material.
WHEN THIS COURSE FITS
Start here when any of these situations is true.
You do not need a level label. Enter when the work shows one of these signals, bring real material, and use course acceptance to decide when you are done.
- 01✓
Answers must show where their support comes from
Policies, research, or product facts influence decisions and readers must inspect consequential claims.
- 02✓
Sources differ by version, region, or permission
The same query can retrieve expired policy, another region’s rule, or restricted material.
- 03✓
The team is considering a RAG investment
You need to decide whether retrieval is warranted and define ownership, citation support, and freshness tests.
COURSE MILESTONES
Every stage leaves a change someone else can review.
These milestones are not another reading list. They show whether the lessons form one piece of work: real material at the start, connected artifacts in the middle, and evidence another person can reproduce at the finish.
- 01 BEFORE↓
Create a small authoritative source set a person can verify.
Bring five questions, including one with no eligible answer. Record source owner, scope, version, effective date, and access permission. Label valid passages before testing retrieval.
Starting evidence: answerable questions map to valid passages and the no-answer case has a stated reason. - 02 MIDPOINT↓
Make the RAG investment decision, then test retrieval.
Compare knowledge change, corpus size, and ownership. If retrieval is warranted, use fixed questions to inspect gold passages, ranking, versions, and access before auditing citations.
Midpoint evidence: each retrieval failure is classified as missing, poorly ranked, stale, unauthorized, or false evidence. - 03 FINISH✓
Challenge answers with conflicting, expired, and unauthorized material.
Ask a reviewer to inspect each consequential claim, confirming that the passage supports it, the version is current, and retrieval has not exposed restricted content.
Completion evidence: bad sources do not win and uncertain answers stop for review.
- 01✓LESSON 1
Prompting, RAG, or fine-tuning? Choose by failure mode
Diagnose the failure first, then compare instructions, request-time retrieval, and fine-tuning on the same holdout cases and operating gates.
LESSON OUTPUT · method-choice trialUnreadOpen lesson → - 02✓LESSON 2
Decide whether RAG is worth building
Compare knowledge volatility, source volume, citation needs, access rules, and maintenance cost before adding retrieval.
LESSON OUTPUT · investment gateUnreadOpen lesson → - 03✓LESSON 3
How do you approve sources before adding them to a RAG knowledge base?
Decide which exact documents and revisions may ground a RAG answer by checking authority, ownership, use rights, audience, effective dates, and conflicts before indexing.
LESSON OUTPUT · rag-source-registerUnreadOpen lesson → - 04✓LESSON 4
How do you validate PDF, table, and image extraction before RAG indexing?
Compare parsed content with original pages and cells before chunking. Check OCR, reading order, table relationships, figures, and source locations.
LESSON OUTPUT · extraction-acceptance-sheetUnreadOpen lesson → - 05✓LESSON 5
How should you chunk RAG documents? Keep the full evidence intact
Compare fixed windows, overlap, and document structure on clauses and tables, then choose chunk boundaries using complete evidence, citations, and cost.
LESSON OUTPUT · document chunking trialUnreadOpen lesson → - 06✓LESSON 6
Keyword, vector, hybrid, or reranking for RAG? Choose with a matched trial
Compare BM25, dense vectors, hybrid fusion, and reranking on exact IDs, paraphrases, and mixed questions while holding evidence, access, context, and latency gates fixed.
LESSON OUTPUT · retrieval strategy comparisonUnreadOpen lesson → - 07✓LESSON 7
Evaluate RAG retrieval before judging the answer
Build a small retrieval trial that checks gold passages, no-answer questions, document versions, ranking, and access boundaries before generation.
LESSON OUTPUT · retrieval trialUnreadOpen lesson → - 08✓LESSON 8
Document-level access control for enterprise RAG, from search to citation
Carry verified identity, tenant scope, and document ACLs into every chunk, then test denial across retrieval, model context, citations, cache, and revocation.
LESSON OUTPUT · document-access traceUnreadOpen lesson → - 09✓LESSON 9
How do you verify that a RAG document update or withdrawal reached the index?
Trace a source revision or withdrawal through ingestion, chunk and vector cleanup, search visibility, caches, and answer probes before calling a RAG knowledge base current.
LESSON OUTPUT · RAG index refresh acceptance trialUnreadOpen lesson → - 10✓LESSON 10
When RAG lacks evidence, should it answer, clarify, or hand off?
Route a RAG request to a supported answer, a specific clarifying question, an honest unable-to-verify response, or human review when retrieved evidence is incomplete.
LESSON OUTPUT · rag-no-evidence-response-gateUnreadOpen lesson → - 11✓LESSON 11
Test citations for support and freshness
Audit whether each important claim is supported by the cited passage, uses the right version, and survives a freshness check.
LESSON OUTPUT · citation auditUnreadOpen lesson →
COURSE ACCEPTANCE
A reviewer can trace every important claim to current, permitted evidence without exposing restricted source material.
Reading is not the finish line. Assemble the lesson outputs and ask another person to review them against the standard above.