--- title: "Devathi Sai Lokesh (@sailokesh8) | Hashnode" description: "AI/ML • MLOps • LLMs | Production Engineering from the Real World. Read the latest articles by Devathi Sai Lokesh on Hashnode." url: "https://hashnode.com/@sailokesh8" --- @sailokesh8 # Devathi Sai Lokesh @sailokesh8 · Bangalore, India · Joined June 2026 AI/ML • MLOps • LLMs | Production Engineering from the Real World ### [4 Layers of Agent Systems: Why Your AI Agent Says It's Done But Actually Failed](https://sailokeshdevathi.hashnode.dev/4-layers-of-agent-systems-why-your-ai-agent-says-it-s-done-but-actually-failed?utm_source=hashnode&utm_medium=feed) Your agent burned through tokens, declared success, and then failed in production. Is it a model problem? Is it a prompting problem? No. It's an architecture problem. The Real Issue When an LLM agent 10 min read · [sailokeshdevathi.hashnode.dev](https://sailokeshdevathi.hashnode.dev/4-layers-of-agent-systems-why-your-ai-agent-says-it-s-done-but-actually-failed?utm_source=hashnode&utm_medium=feed) [Aug 15](https://hashnode.com/posts/4-layers-of-agent-systems-why-your-ai-agent-says-it-s-done-but-actually-failed/6a8077cbf9593b0ad69a8bcc) ### [AI Governance in Production: How to Stop Agents from Overspending, Leaking Data, and Calling Tools They Shouldn't](https://sailokeshdevathi.hashnode.dev/ai-governance-in-production-how-to-stop-agents-from-overspending-leaking-data-and-calling-tools-they-shouldn-t?utm_source=hashnode&utm_medium=feed) You have 10 different AI models, 15 custom agents, 8 external tools, and 0 visibility. Here's how to fix it. The Problem: Distributed AI Chaos Your company is deploying AI everywhere: Models: OpenAI 12 min read · [sailokeshdevathi.hashnode.dev](https://sailokeshdevathi.hashnode.dev/ai-governance-in-production-how-to-stop-agents-from-overspending-leaking-data-and-calling-tools-they-shouldn-t?utm_source=hashnode&utm_medium=feed) [Jul 31](https://hashnode.com/posts/ai-governance-in-production-how-to-stop-agents-from-overspending-leaking-data-and-calling-tools-they-shouldn-t/6a6cd62772e201723cbf3b1e) ### [Why Your LLM Times Out in Production and How to Fix It](https://sailokeshdevathi.hashnode.dev/why-your-llm-times-out-in-production-and-how-to-fix-it?utm_source=hashnode&utm_medium=feed) LLM worked perfectly in testing. Shipped it to production. Now users are complaining it takes 8 seconds to respond. Some requests are timing out. Monitoring shows high latency but no idea where to sta 7 min read · [sailokeshdevathi.hashnode.dev](https://sailokeshdevathi.hashnode.dev/why-your-llm-times-out-in-production-and-how-to-fix-it?utm_source=hashnode&utm_medium=feed) [Jul 21](https://hashnode.com/posts/why-your-llm-times-out-in-production-and-how-to-fix-it/6a5f87fe7ad84da66be87564) ### [How to Evaluate LLM Agents in Production](https://sailokeshdevathi.hashnode.dev/how-to-evaluate-llm-agents-in-production?utm_source=hashnode&utm_medium=feed) Evaluating a model is simple. Give it an input, check the output, measure accuracy. Evaluating an agent is completely different. An agent does not produce one output. It makes a plan, calls tools, rea 13 min read · [sailokeshdevathi.hashnode.dev](https://sailokeshdevathi.hashnode.dev/how-to-evaluate-llm-agents-in-production?utm_source=hashnode&utm_medium=feed) [Jul 12](https://hashnode.com/posts/how-to-evaluate-llm-agents-in-production/6a5337d5aa06f71aca6c1307) ### [MCP (Model Context Protocol) — How AI Agents Finally Got a Standard Way to Use Tools](https://sailokeshdevathi.hashnode.dev/mcp-model-context-protocol-how-ai-agents-finally-got-a-standard-way-to-use-tools?utm_source=hashnode&utm_medium=feed) Every time your agent needs a new tool, someone has to write custom code to connect them. GitHub integration — custom code. Slack integration — custom code. PostgreSQL integration — custom code. File 9 min read · [sailokeshdevathi.hashnode.dev](https://sailokeshdevathi.hashnode.dev/mcp-model-context-protocol-how-ai-agents-finally-got-a-standard-way-to-use-tools?utm_source=hashnode&utm_medium=feed) [Jul 5](https://hashnode.com/posts/mcp-model-context-protocol-how-ai-agents-finally-got-a-standard-way-to-use-tools/6a4a42b71b2e6585409d35c1)