Andy Wong

This open source RAG approach is amazing. The entire RAG industry may get cooked 🔥 This thing: - runs without a vector database. - skips embeddings entirely. - doesn't chunk your source document. - ignores similarity search completely. It's unlike everything else you're seeing on the market, and it just scored 98.7% on FinanceBench, which tests long financial filings. I've been building RAG stacks for businesses with vector retrieval. Most of them still crack at the seams with long docs and videos. PageIndex has a different approach, and it's open source and free: github.com/VectifyAI/PageInd… Here's what you need to know: 1️⃣ The index is a tree ↳ Instead of a vector index, it generates a table of contents for your PDFs ↳ Then an agent navigates it, like AlphaGo's tree search ↳ Retrieval returns references instead of vector hits 2️⃣ Why it beats chunked vector RAG ↳ Similarity is not the same as relevance ↳ Instead of vibe retrieval, every retrieval is traceable 3️⃣ What you don't have to run anymore ↳ No vector database to host ↳ No embeddings to rebuild every time your docs update ↳ A simpler stack that carries conversation context across turns 4️⃣ How to try it today ↳ Tell Claude Code or Codex to look at the GitHub repo ↳ Then have it build it in a secure, scalable way Realistically, vector RAG is far from dead. But this is the architecture I'd reach for in many long professional document cases. My vector database took the news poorly. How's your RAG doing? Let me know. ♻️ Repost to help other builders in your network. And follow Aakash Gupta for more on building with AI. 📷 Thank you, Rakesh Gohel, for the infographic. Follow him! ====== BONUS: 5 resources to go further on this: 1. RAG vs fine-tuning: news.aakashg.com/p/rag-vs-fi… 2. AI basics: news.aakashg.com/p/ai-founda… 3. Context engineering: news.aakashg.com/p/context-e… 4. Harness engineering: news.aakashg.com/p/harness-e… 5. Company OS: news.aakashg.com/p/company-o…