Archives

Signing for a faster model without the rest of the system is like signing a star player and playing without fielders or DRS. A breakdown of the six layers every production AI agent needs.

20 open-source OCR and PDF extractors for RAG and AI agents, and why the best OCR model isn't the best pipeline.

Pune's Ganesh Utsav is easy to celebrate, but not always easy to navigate. So I built a simple, ad-free guide to help people find Ganpatis, aartis, and routes, right when they need them.

RAG is more than chunking, embedding, retrieval, and generation. See the steps in between: parsing, reranking, and relevance checks, and where each one fails.

Why AI gets letter counts wrong and what it tells us about artificial vs human brains.

Bad chunks lose answers. This post covers why agentic chunking outperforms traditional methods, and a simple corn test that shows it.

Still using basic PDF parsers? Docling adds layout detection, table reconstruction, and semantic chunking for modern LLM pipelines.

Most RAG pipelines fail because of bad chunking, not the model. This post covers how semantic chunking, splitting text by meaning instead of size, improves retrieval accuracy.

Improve RAG retrieval with the right chunking strategy. Compare fixed-size, recursive, document-specific, semantic, and agentic chunking, including when each one fails and how to choose.

Complete guide to connecting FastAPI and Google Drive using OAuth2

Use Ed25519 signatures to issue tamper-proof, copy-pasteable licenses that your app verifies locally.

Verifying that a user's license is valid and hasn't been tampered with is a hard problem. This tutorial builds a text-based signing and verification system around RSA, which remains the long-standing standard for digital signatures.