Most enterprise AI projects do not fail at the model. They fail in the six months after the demo - evals nobody trusts, retrieval that degrades under real permissions, agent harnesses that cost more than they save, and an infrastructure bill that arrives before the ROI does. That gap is what I work on. Over 15 years I have architected the systems underneath a 22-language production translation platform at Devnagri, high-traffic logistics and e-commerce infrastructure at Shiprocket, and enterprise delivery at ValueLabs. One platform overhaul I led drove 8x ARR growth while cloud spend went down, not up. What I am usually brought in for: -> An LLM or agent system that works in a notebook and falls over in production -> RAG that returns the wrong thing once real tenancy and permissions are applied -> Eval and LLMOps strategy, so you know it is working after you ship it -> Multilingual and voice pipelines, especially Indic languages, ASR and OCR -> Cloud architecture across AWS, Azure and GCP where the cost model is the constraint -> A second opinion before a board or an investor asks the question you cannot answer yet I am direct about tradeoffs and I will tell you when the answer is that you do not need the thing you came to buy. Send me the diagram for a written second opinion, or book the hour if you would rather talk it through.