More of what I've written

All Articles

Writing on agentic AI and LLM application development — practical lessons from building, deploying, and streaming responses from real AI agents.

  1. 5 Guardrails I Add Before Any AI Agent Touches Real Data cover image

    5 Guardrails I Add Before Any AI Agent Touches Real Data

    Medium

    Five infrastructure-level safety layers for agents with real-world access: scoped tool permissions, allow-listed destinations, data-loss prevention, human approval gates, and full action logging.

  2. From Prototype to Production: Lessons from Deploying My First Agent cover image

    From Prototype to Production: Lessons from Deploying My First Agent

    Medium

    What it actually takes to ship an agent to production: background workers, Docker, env-based config, timeouts/retries on external calls, and full execution-trace logging.

  3. Streaming Agent Responses to a React/Next.js UI in Real Time cover image

    Streaming Agent Responses to a React/Next.js UI in Real Time

    Medium

    Streaming an agent's progress — text, tool status, completion — to a React UI in real time using Server-Sent Events and a FastAPI backend.

  4. Single-Agent vs Multi-Agent Systems: When Do You Actually Need More Than One? cover image

    Single-Agent vs Multi-Agent Systems: When Do You Actually Need More Than One?

    Medium

    Why a single well-equipped agent usually beats multi-agent orchestration — the latter only pays off on broad, parallel tasks, at roughly 3.75x the token cost.

  5. Short-Term vs Long-Term Memory in AI Agents: A Practical Implementation cover image

    Short-Term vs Long-Term Memory in AI Agents: A Practical Implementation

    Medium

    A dual-memory architecture with PostgreSQL + pgvector — rolling-summary short-term memory plus embedding-based long-term memory — to keep prompt size constant as conversations grow.