VKVamsi Krishnainprodainotes.hashnode.dev·1d ago · 3 min readDay 11: Why RAG is an evidence design problem, not a buzzwordRAG is often explained as letting the model look things up. That is true, but the production question is sharper: what evidence should the system trust before it answers? A useful RAG system is not ju00
VKVamsi Krishnainprodainotes.hashnode.dev·Jun 30 · 3 min readDay 10: When to use a chatbot, workflow, or agentNot every AI feature should be a chatbot. That sounds obvious until every prototype starts with a blank text box. Chatbots, workflows, and agents solve different problems. A chatbot is good for conver00
VKVamsi Krishnainprodainotes.hashnode.dev·Jun 30 · 3 min readDay 09: Why a capable model is still not a productA capable model can still produce a weak product. The gap is usually not intelligence; it is product design. Users do not experience benchmark scores. They experience onboarding, defaults, latency, er00
VKVamsi Krishnainprodainotes.hashnode.dev·Jun 30 · 3 min readDay 08: Why production AI needs structured outputsFree-form text is comfortable for humans and awkward for systems. Production software usually needs something stricter. Structured output turns a model response into something downstream code can vali00
VKVamsi Krishnainprodainotes.hashnode.dev·Jun 30 · 3 min readDay 07: Why confident AI answers still need evidenceThe hardest hallucinations are not the obviously strange ones. The hardest ones are fluent, confident, and almost right. A production system cannot treat confidence in language as confidence in truth.00
VKVamsi Krishnainprodainotes.hashnode.dev·Jun 30 · 3 min readDay 06: Why creativity settings are product decisionsTemperature is often treated like a creativity knob. That is true, but incomplete. In a product, sampling settings decide how much variation users will experience. A writing assistant, a support workf00
VIVictor Isuoinvictorisuo.hashnode.dev·Jun 26 · 10 min readWhat Mechanical Engineering Taught Me About Building AI SystemsThe third agent was the one that broke everything. It was 11pm. I was debugging LexAI — a four-agent CrewAI contract analysis pipeline — and the system was collapsing at exactly the same point on ever00
VKVamsi Krishnainprodainotes.hashnode.dev·Jun 26 · 3 min readDay 01: Why model choice is rarely the first production AI problemA lot of AI projects begin with the same question: which model should we use? It is a reasonable question, but it is rarely the first one that matters in production. The model is only one component in02M
VKVamsi Krishnainprodainotes.hashnode.dev·Jun 26 · 3 min readDay 05: Why vague prompts become vague systemsA vague prompt can still produce a good-looking answer in a demo. That is exactly why it is dangerous. Production prompts carry expectations: tone, task, boundaries, output shape, refusal behavior, an00
VKVamsi Krishnainprodainotes.hashnode.dev·Jun 26 · 3 min readDay 02: Why LLMs feel intelligent but still generate one step at a timeLLMs can feel like they understand a whole answer before they begin writing it. Under the hood, the behavior is much more incremental. The system is choosing the next likely token again and again. Tha00
