MKMudassir Khaninmudassirworks.hashnode.dev·1d ago · 7 min readVector Search Quantization: How to Keep Your RAG Pipeline Fast and CheapIf you have shipped a retrieval augmented generation system past a demo and into production, you already know the pain. Your vector index started small and fast, then your dataset grew from a hundred 00
MKMudassir Khaninmudassirworks.hashnode.dev·Jul 10 · 9 min readLLM Evaluation Metrics: What Actually Predicts Production QualityYou shipped a RAG pipeline that scores well on the vibes test. Answers look reasonable, your team is happy, and the demo went great. Then production happens: users start hitting edge cases your gut te00
MKMudassir Khaninmudassirworks.hashnode.dev·Jul 7 · 9 min readHybrid Search for RAG: Combining BM25 and Vector Search for Better RetrievalIf you have ever watched a RAG pipeline confidently return the wrong chunk, there is a good chance the problem was not your prompt or your model. It was your retriever. Vector search alone has a speci11K
MKMudassir Khaninmudassirworks.hashnode.dev·Jul 2 · 7 min readRealtime AI Streaming in Next.js: SSE vs WebSockets vs Vercel AI SDKYou're building an AI chat feature and you've already wired up the API client. The first call works, but the response takes several seconds to appear all at once. Users see a blank screen, then a wall00
MKMudassir Khaninmudassirworks.hashnode.dev·Jun 28 · 10 min readNext.js Authentication Security: Patterns That Hold Up in ProductionNext.js authentication security is one of those topics where a tutorial gets you 80% of the way there, and the remaining 20% is where real applications fail in production. This article covers the secu00
