Writing on AI systems, cloud & infra
RSS

Kevin Tan

Cloud Architecture · AI Engineering · Distributed Systems

Latest · Post

Cross-Document Retrieval for AI Agents Without a Vector Database

BM25 scores don't compare across documents. Ranks do. Fusing 100 PDFs into one hit list.

Aerial view of a braided glacial river: independent channels merging into one flow, the way per-document rankings fuse into one cross-corpus result. Ranks survive that merge; raw BM25 scores do not.

Most read guides — reader favorites

View archive →

More Articles

01

Why Multi-Column PDFs Scramble Reading Order in RAG

July 18, 2026 8 min

02

How to Evaluate LLM Agents: 5 Patterns That Catch Failures

July 11, 2026 12 min

03

RAG for AI Agents: 6 Decisions That Make or Break Retrieval

July 10, 2026 10 min

04

How One Search Change Eliminated an Entire Agent Step

July 04, 2026 8 min

05

How to Evaluate an MCP Server With an LLM: 17 Bugs Found and Fixed

June 27, 2026 12 min

Highlight Editor's pick →

The Dispatch

Engineering essays from real production work, in your inbox.

AI agents, cloud architecture, reliability, and the trade-offs behind them. Roughly weekly.

No spam. Unsubscribe in one click.