CIAMETECHLoading
Working notes on agentic AI, cost engineering, architecture, and building production AI for India. Capability and method, never invented results.
AI for India has to begin where the country actually is: many languages, patchy bandwidth, and real rules about where data can live. What that means in practice, for business and for government.
ReadPublic-sector AI fails in predictable ways: pilots that never scale, tools no one trusts. What it takes to get from a problem brief to something citizens and officers genuinely use.
ReadThree approaches get mixed up constantly, and picking the wrong one is expensive. A straight explanation of what RAG, fine-tuning, and small models each actually do, and when to reach for which.
ReadYour AI bill is mostly an engineering choice, not a fixed cost. Four levers, tokens, caching, model routing, and visibility, that bring it down while users notice nothing except speed.
ReadA chatbot answers a question and waits. An agent gets the job done. Here is what changes when software can plan, act, and check its own work, and where it pays off first.
Read