・Build the backend services behind conversational and autonomous commerce agents: intent understanding, multi-turn task execution, tool orchestration, human confirmation checkpoints

・Own the commerce tool layer - catalog search, product detail, inventory and price lookup, cart, checkout, order status, returns - exposing internal commerce APIs to agents with strict contracts, auth scoping, and blast-radius limits

・Solve grounding on a live catalog: retrieval over tens of millions of SKUs, attribute and variant normalization, freshness guarantees on price and stock, and preventing agents from confidently recommending things that don't exist or can't ship

・Design transactional safety for agent actions: idempotency, confirmation and rollback flows, spend and scope limits, permission models, and full audit trails for every action an agent takes on a user's or merchant's behalf

・Build evaluation and observability for commerce outcomes - task success, recommendation relevance, cart and conversion impact, containment and escalation rates - with tracing across multi-step runs and regression gates before release

・Optimize latency and unit economics: retrieval and tool-call budgets, model routing, prompt caching, cost per resolved task measured against the revenue or support cost it moves

・Work directly with commerce product, search/recsys, merchant platform, and trust & safety teams to ship agents through launch review into production

Mandatory Qualifications:

・Over 5 years building and operating production backend services; strong in Python (Go or Java a plus)

・Experience developing ML-based production systems - search, recommendation, ads ranking, personalization, or fraud/risk. You've owned a system where model quality and system quality were the same problem

・Experience with high-traffic transactional or e-commerce backends: catalog, inventory, cart/checkout, payments, or order management - you understand why correctness and consistency matter more here than in most systems

・Track record of shipping quality improvements measured by data: offline evaluation, online A/B testing, and the judgment to know when they disagree and why

・Solid distributed systems fundamentals: async and event-driven architecture, queues, concurrency, idempotency, distributed tracing

・Hands-on experience shipping LLM-backed systems to production - not just API calls, but the infrastructure around them

・Direct exposure to agentic patterns: tool/function calling, orchestration frameworks, multi-turn state management

・Cloud and containerized deployment experience (Kubernetes, CI/CD, IaC)

#engineer #applicationsengineer #aianddatadiv