Skip to content

Chapter 16: Retrieval-Augmented and Knowledge-Driven Agents

Writing scaffold: This page fixes the chapter path and delivery fields. The responsible student will replace it with reviewed English prose after the Chinese manuscript is frozen.

Field Current baseline
Owner D
Target equivalent pages 36
Part Part 4: Memory, Knowledge, and Agent Learning
Dependencies Ch15 → this chapter → Ch17
Version status en-scaffold-0.1

1. Chapter role

This chapter develops verifiable concepts, engineering boundaries, and reader tasks for “Retrieval-Augmented and Knowledge-Driven Agents”. The final manuscript should begin with a concrete failure scenario and explain the problem, method, trade-offs, evidence, and limits.

2. Learning outcomes

By the end of this chapter, readers should be able to:

  1. Explain the central concepts, inputs, outputs, and assumptions.
  2. Design a minimal solution compatible with the shared book contracts.
  3. Identify failure signals, validation evidence, and human handoff boundaries.

3. Suggested structure

  1. Problem scenario and failure mode
  2. Core concepts, model, or system boundary
  3. Engineering workflow, interfaces, and trade-offs
  4. Case study, experiment, and limitations
  5. Summary, review questions, and practical exercise

4. Figure and table placeholders

  • Fig-16-01: core architecture or workflow (source and license to be registered).
  • Fig-16-02: failure path or validation workflow.
  • Tab-16-01: terminology, interface, or metric comparison.
  • Tab-16-02: case, experiment, or acceptance matrix.

5. Code and experiment

  • Experiment card: Exp-16-01, including question, hypothesis, data/license, versions, control, repetitions, success criteria, and raw results.
  • Code entry point: examples/ch16/ (to be created).
  • Reproduction command, expected output, and known limitations: pending.

6. Delivery checklist

  • English prose is near the target page budget and covers all six structural blocks.
  • Figures, citations, terminology, and cross-chapter links use shared IDs.
  • Case studies, simulated results, and real measurements are clearly separated.
  • Review questions, practical exercise, and scoring guidance are complete.

References and version record

The final manuscript records at least eight key sources and keeps claim, source location, applicability conditions, and verifier in claim_sources.md.

  • Current version: en-scaffold-0.1
  • Last updated: 2026-09-09