Lecture 04. Retrieval as a tested subsystem

Part 2 · The Agent as Tested Subsystems · Scheduled 29 Nov 2026 · 1-hour lecture

Upcoming course outline. This class has not been recorded as studied.

Learning objectives

• Chunking; BM25; dense retrieval; hybrid RRF; reranking.
• Metrics: Recall@k, MRR, nDCG, citation precision; how to label a query set.
• Failures: recall ceiling, stale index, permission leak, lost in the middle, injection through documents.

Why this class matters

• Industry: RAG is the most common agent feature in Bangladeshi and foreign companies; students ship it with a metric and a permission test.
• Academic: Bangla retrieval on a bilingual corpus is under-studied; the metric table is a paper section.
• Elsewhere: Students can tell a 'search' demo from a measured retrieval system.

Preparation requirements

• Stanford CS329Z L3 and L4 slides (complete).
• The paper (Lewis et al. 2020) as pre-reading.

These are preparation requirements in the draft; staff materials are not bundled with this site.

Preparation and readings

• Stanford CS329Z L3: Retrieval-augmented generation (slides)
• Stanford CS329Z L4: Tool use and function calling (slides)
• Lewis et al., RAG (NeurIPS 2020)

The linked reference library contains the source URLs extracted from the course PDF. Availability notes above reflect the draft of 4 November 2026.

Paired lab

Continue with Lab 04: Retrieval as a tested subsystem. Every lab contributes an artifact to the same repository and a measurement to the results ledger.

Personal notes

[Lecture notes and observations]