https://pranavgore.com/series/building-llm-agents-in-production/ Recent content in Building LLM Agents in Production on Pranav Gore | Staff Backend Engineer https://pranavgore.com/img/og_v1.jpg https://pranavgore.com/img/og_v1.jpg Hugo en Fri, 24 Jul 2026 10:00:00 +0800 https://pranavgore.com/blogs/evaluating-llm-agents-before-eval-infrastructure/ Fri, 24 Jul 2026 10:00:00 +0800 https://pranavgore.com/blogs/evaluating-llm-agents-before-eval-infrastructure/ Why our first eval for an LLM agent is 90 lines, four scripted scenarios, and a human judge instead of an eval framework, and when that stops being enough. https://pranavgore.com/blogs/llm-in-production-structure-cost-defenses/ Sat, 18 Jul 2026 10:00:00 +0800 https://pranavgore.com/blogs/llm-in-production-structure-cost-defenses/ Wrapping an untrusted LLM for production: structured output, completion the model can't fake, a reliability sandwich, cost per conversation, and prompt-injection defenses. https://pranavgore.com/blogs/conversational-ai-agents-reply-to-conversations/ Thu, 09 Jul 2026 10:00:00 +0800 https://pranavgore.com/blogs/conversational-ai-agents-reply-to-conversations/ Why a chat agent should attach to a conversation and poll its state instead of replying per message, and how it decides when to speak, wait, and stop. https://pranavgore.com/blogs/durable-agent-runs-postgres-celery/ Fri, 03 Jul 2026 10:00:00 +0800 https://pranavgore.com/blogs/durable-agent-runs-postgres-celery/ How we built durable, crash-safe, long-running conversational agents on a single Postgres table and Celery, without a workflow engine like Temporal.