DSDevathi Sai Lokeshinsailokeshdevathi.hashnode.dev·2d ago · 13 min readHow to Evaluate LLM Agents in ProductionEvaluating 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, rea00
DSDevathi Sai Lokeshinsailokeshdevathi.hashnode.dev·Jul 5 · 9 min readMCP (Model Context Protocol) — How AI Agents Finally Got a Standard Way to Use ToolsEvery 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 00
DSDevathi Sai Lokeshinsailokeshdevathi.hashnode.dev·Jun 25 · 8 min readContext Engineering is Not Prompt EngineeringEveryone optimized their prompts. The agents still gave wrong answers. The issue was never the instruction. The issue was everything around it — what the model was allowed to see, what it was not, how00
DSDevathi Sai Lokeshinsailokeshdevathi.hashnode.dev·Jun 14 · 10 min readThe 5 Layers of Agent Memory — What Every Production Agent NeedsEveryone talks about context engineering. Nobody shows you the memory stack underneath it. Without memory, an agent forgets everything after each session. Like talking to someone with amnesia — you sh00
DSDevathi Sai Lokeshinsailokeshdevathi.hashnode.dev·Jun 12 · 8 min read# Not Every RAG System Needs a Vector DatabaseEveryone building RAG systems starts the same way. Document → Chunks → Embeddings → Vector Database → Similarity Search → LLM That pipeline works. But it is not the only way to retrieve information. A00
