Applied AI
Designing Resilient Schema Guardrails for Applied LLM Workflows
Why prompt engineering alone fails in enterprise production, and how strict Pydantic/Zod validation layers enforce deterministic outputs.
Essays & Technical Notes
Writing
I write about applied AI, software engineering, and the craft of building products that work for real users. No fluff — just lessons from the work.
Engineering
From Client Calls to Code: How Full-Stack Literacy Cuts Handoff Friction
When engineers understand commercial context and customer friction directly, PRD specifications shrink and sprint velocity multiplies.
AI Engineering
Why Your Chatbot Doesn’t Need GPT-4 for Routing: The Case for Self-Trained SLMs
Relying on frontier LLMs to categorize user intent adds latency, ballooning token bills, and cloud vendor lock-in. A fine-tuned Small Language Model (SLM) trained for under $10 solves query routing with sub-50ms deterministic speed.
Architecture
Building ConnexxIQ: Scaling Read-Replica Queries for Real-Time Analytics
Architectural lessons from isolating customer engagement reporting without table locks or latency penalties on core production databases.
Systems
Why Every AI Agent in Production Needs a Human-in-the-Loop Override
Autonomous workflows are only as dependable as their rollback controls. Designing clean audit traces and operator verification gates.