Getting a demo working is easy. Keeping an LLM feature reliable, safe, and affordable with real users is the actual engineering. That’s what I do here.
Production LLM engineering
- Evaluation — automated eval suites so you can change prompts/models without guessing whether quality moved.
- Monitoring & observability — track quality, latency, cost, and failures in production.
- Prompt & version control — manage prompts and model routing (e.g., via OpenRouter) so you can swap models without code rewrites.
- Cost & latency management — right-size models and caching to keep spend and response times in check.
- LLM security — prompt-injection defense, output validation, and safe handling of tools and sensitive data.
AI for E-commerce
Where it’s a fit, I apply production AI to commerce: product/catalog RAG and search, customer-support and operations automation, and content workflows — backed by years of hands-on e-commerce delivery. (Legacy Magento work is archived; the focus now is LLM-powered commerce.)