MKMudassir Khaninmudassirworks.hashnode.dev·16h ago · 9 min readLLM OCR for RAG: A Document Parsing Pipeline Guide That Protects Retrieval QualityIf you searched ocr llm because your retrieval augmented generation system keeps returning garbled answers from scanned PDFs, the problem almost never starts at the language model. It starts at the pa00
MKMudassir Khaninmudassirworks.hashnode.dev·1d ago · 7 min readAgentic Search vs Traditional RAG: When One Query Is Not EnoughAgentic search is retrieval where an LLM plans the search instead of firing one embedding lookup. It decomposes a complex question into focused subqueries, runs them in parallel against your knowledge10
MKMudassir Khaninmudassirworks.hashnode.dev·5d ago · 10 min readStructuring Next.js AI Apps for Scale: Architecture Patterns That Hold Under Production LoadBy the end of this article, you will have a working mental model for nextjs ai app architecture: where LLM calls belong in the App Router, how to keep streaming responses fast without fighting React, 00
MKMudassir Khaninmudassirworks.hashnode.dev·Jul 31 · 13 min readLLM Knowledge Leakage Prevention: Retrieval Permissions, Gateway Scrubbing, and Audit TrailsYour access control is probably correct and your model is probably still leaking. That is the uncomfortable part of llm knowledge leakage prevention: the permission check runs at the document store, a00
MKMudassir Khaninmudassirworks.hashnode.dev·Jul 29 · 9 min readTypeScript Design Patterns for LLM Applications: From Prototype to ProductionBy the end of this article you will be able to structure a TypeScript AI application architecture that survives contact with production: a swappable LLM layer, typed tool calling, streaming responses,00
