Andrey Shkanov

I build, not delegate. I embed, scope the workflow, build it, ship it to production, and transfer the capability to your team — the models mostly work; value leaks at deployment, so that's where I live.

• Production LLM work with real prompt engineering — not "I've used an LLM."
• Agent frameworks and tool-use chains — multi-step workflows that call APIs and write to real systems.
• Evaluation and cost-aware model routing — the right model per task, proven against production workloads.
• Local inference and homelab infrastructure — on-prem and multi-node Apple-Silicon inference.
• Problem decomposition and systems thinking — scoping a workflow correctly before shipping it.
• End-to-end ownership — discovery → production → a measured outcome, transferred to your team.
• Agentic, autonomous workflow design — building the agent-driven automation customer-side.

Forward-deployed and applied-AI work spanning enterprise migrations and incident leadership, plus healthcare and medical-device contexts.

• Enterprise migration under outage pressure → service restored in under ~45 minutes (disciplined incident trace + misconfiguration isolation).
• Post–go-live stabilization → restored order processing and stayed engaged ~6 weeks across NA + EU teams to rebuild confidence and prevent churn.
• Product influence → structured customer feedback surfaced workflow ambiguity driving overblocking; partnered with Product to reduce operational risk.
• Executive communication → led Application Security workshop at Cloudflare Connect (Cloudflare University) focused on operational implications and trade-offs.