~/series/
Data & Retrieval Engineering in 30 Days
30 lessons published · new lesson added daily.
1
Day 1
Why Bad Input Data Sinks Agentic Coding Tasks
2
Day 2
Multi-Agent Pipelines Are ELT, and the Transform Step Goes Unreviewed
3
Day 3
When streaming with Kafka isn't worth it
4
Day 4
Why an agent doesn't know which data table to trust
5
Day 5
How an agent's dbt refactor silently inflated revenue
6
Day 6
Why a lakehouse migration doesn't change schema ownership
7
Day 7
Why agents guess field meanings instead of asking
8
Day 8
How an AI agent overloaded a partition key that worked fine
9
Day 9
Why Parquet queries are faster than CSV (not compression)
10
Day 10
The dual-write bug in event-driven systems, and how to fix it
11
Day 11
How a Self-Retrying Payment Pipeline Caused a Double Charge
12
Day 12
How a Cache Made an Agent Think It Was Still March
13
Day 13
Why Stale Data Is More Dangerous Than No Data for Agents
14
Day 14
How out-of-order backfill makes old data look current in your index
15
Day 15
How embedding drift degrades RAG when your corpus shifts
16
Day 16
Why swapping BM25 for a vector DB can silently lower recall
17
Day 17
Why bad chunking breaks RAG before embeddings can help
18
Day 18
How a cross-encoder reranker fixes RAG relevance without re-embedding
19
Day 19
Hybrid Search Fusion Tuning
20
Day 20
Semantic Search
21
Day 21
Chunking failures in RAG
22
Day 22
Knowledge decay
23
Day 23
Lexical search vs embeddings for code
24
Day 24
Context precision
25
Day 25
Retrieval eval sets
26
Day 26
Semantic data contracts
27
Day 27
Agent memory as a feature store
28
Day 28
Agent dead letter queues
29
Day 29
Per-agent context tax
30
Day 30
Lineage for agentic pipelines