RAG & Data Pipelines
Build and troubleshoot retrieval-grounded applications.
Trace a document from upload to cited answer. Diagnose missing evidence stage by stage, and design a multi-tenant pipeline that stays correct through updates, retries, and deletion.
6 questions
Questions
Easy
1.5 h
Explain how documents become evidence for a generated answer, and why an answer can be wrong even when it cites a real document.
1.5 h
Turn a follow-up message into a standalone retrieval query, keep access filters tied to the signed-in session on every turn, and decide which turns should not retrieve at all.
Medium
3 h
Trace a missing policy exception through ingestion, eligibility, and context packing without weakening access rules.
3 h
Design a support assistant that answers from help articles with citations the customer may see, or hands off to a human. Turn its requirements into metrics, plan its evaluation set, keep it fresh through updates and deletion, and check its token, cost, and latency budgets.
Hard
4 h
Design a private multi-tenant policy assistant whose answers and citations stay correct through asynchronous updates, retries, permission changes, and deletion.
4 h
Find why a partly failed re-embedding backfill left a help-center index missing, duplicating, and mixing vectors, then plan a repair and a cutover gate you could defend in a release review.
What you will be able to do
- Distinguish retrieval from generation
- Explain the ingestion and query paths
- Identify the purpose of chunks and metadata
- Recognize unsupported answers even when they contain citations
- Explain why a follow-up message retrieves the wrong evidence, or none, when it is sent to retrieval as typed
- Rewrite a follow-up into a standalone retrieval query, keep the original message for the generator, and log both
- State where access filters come from on every turn, and why neither the user's text nor a rewrite can widen them
- Decide when a turn should skip retrieval, ask a clarifying question, or start a fresh query after a topic switch
- Name the ways a rewrite can fail and the few multi-turn test cases that catch them
Hands-on exercises
Interview rounds that use this topic
Topic boundaries
This topic covers ingestion, grounding, citations, and the document lifecycle. Embeddings & Retrieval covers how candidates are found and ranked.
Related on CodeSnatch
System Design
Covers idempotent tasks, safe backfills, and quality checks that keep a document ingestion pipeline recoverable.
Community article
A practitioner's tour from ingestion to cited answer, with real failure modes such as stale corpora and invented citations.
System Design
Shows how an indexing pipeline handles deletes, refresh, and reindexing behind an alias, the mechanics behind document updates.
System Design
Covers multi-tenant authorization and row-level security, the controls that decide which chunks a user may retrieve.
System Design
Compares TTL, write-driven, and event-driven invalidation, which cached answers need when a source changes or is deleted.
AI Engineering
The natural next step: read a request trace to separate retrieval, packing, and generation failures.
AI Engineering
Implements a context selector that enforces tenant, group, and deletion rules, then proves nothing forbidden leaks.
System Design
Explains at-least-once delivery and idempotent consumers, the basis for ingestion jobs that retry safely.
System Design
A worked ingestion system with parsing, content deduplication, and recrawl scheduling that keeps a corpus fresh.