Archive - AI Exploration Journey

Attention Is All You Need—for Retrieval Too? — AI Innovations and Insights 141

Wait, is external retrieval in RAG actually redundant?

Jun 19

MinerU-Popo: Fixing the Hidden Flaw in PDF Parsing — AI Innovations and Insights 140

Modern OCR can produce impressive page-level results, but real documents do not think in pages. They think in chapters, sections, explanations, figures…

Jun 14

Unlocking PDFs for RAG: How RAG-Anything Handles Complex Documents — AI Innovations and Insights 139

PDF files are notoriously complex.

Jun 10

The Hidden Bottleneck in RAG Code Completion: How You Chunk the Code — AI Innovations and Insights 138

In a previous article, I discussed how Claude Code handles retrieval (The Secret Behind Claude Code’s Retrieval: Why Live Search Fits Better than RAG …

Jun 6

Building Better LLM Agents with Smarter Memory Architectures — AI Innovations and Insights 137

If you’ve ever used an AI assistant for a long project, you’ve probably felt the frustration: it is brilliant one moment and forgetful the next.

Jun 2

May 2026

Can We Build an Atlas for AI Research? — AI Innovations and Insights 136

Every researcher knows the strange feeling of reading a paper and realizing there are ten earlier papers you probably should have read first.

May 29

DocSeeker: Page by Page, Evidence by Evidence — AI Innovations and Insights 135

If you have ever asked a model a question about a long report and received a confident but unverifiable answer, you know the frustration.

May 25

The Secret Behind Claude Code's Retrieval: Why Live Search Fits Better than RAG — AI Innovations and Insights 134

In AI coding tools, finding code isn’t just about a search box.

May 21 • Florian

Corpus2Skill: Turning Flat RAG into Browsable Agent Skill Trees — AI Innovations and Insights 133

Agent Skills have become a popular technique recently (AI Exploration Journey: Skills).

May 17

SkillClaw: One Agent Fails, Every Agent Learns — AI Innovations and Insights 132

Agent skills evolving is currently an exciting new direction.

May 12

MinerU-Diffusion: A New Path Beyond Autoregressive OCR — AI Innovations and Insights 131

The uncomfortable truth is that some OCR systems look smarter than they are because language helps them fill in the blanks. But when the page stops…

May 8

MultiDocFusion: From Flat Chunks to Hierarchy-Aware RAG — AI Innovations and Insights 130

For RAG, a lot of us begin with the same almost invisible premise: chunk the document, embed the chunks, retrieve the top matches, and the rest will…

May 4