Machine Learning Engineer III

The Company CommerceIQ is building the AI platform that runs commerce for the world's largest brands. We are not selling AI demos. We are shipping AI agents for content, media, and sales into the workflows of the Fortune 100 every week. 2,200+ Customers 10 of Top 12 CPG Companies 900+ Retailers Connected $200M+ Raised Customers include Coca-Cola, Nestlé, Colgate-Palmolive, Mondelez, Samsung, and Kellogg's. Backed by SoftBank, Insight Partners, and Madrona. Headquartered in Mountain View with teams across the US, India, Canada, and the UK. Pre-IPO. About the Role We are looking for a hands-on Machine Learning Engineer III to build and run the ML and AI systems behind our agents for content, media, and sales. You will work closely with Data Science, Engineering, Platform, Product, and Operations teams to take models and LLM-based agents from idea to production, and keep them reliable, fast, and cost-efficient at scale. This is an engineering role first: you will own your systems end to end, including the infrastructure they run on. We look for builders who ship v0s quickly, iterate against evals, and stay adaptable about models and frameworks as the field moves. Key Responsibilities : AI Agents & Harnesses: Design, build, and ship production agents and the harnesses around them: tool and action design, context window management (retrieval, summarization, compaction), model selection and routing, memory and state, and subagent and parallel tool-call patterns. Iterate until agents reliably complete real customer tasks. ML Pipelines: Build end-to-end batch and real-time pipelines for classical ML and LLM workloads, from data ingestion to deployment. Infrastructure: Own the infrastructure your systems run on: containers, Kubernetes, queues, caches, and cloud services on AWS/GCP, managed as code. Reliability & Quality: Meet latency, throughput, and cost budgets using caching, batching, and parallel execution. Handle rate limits, retries, and failures gracefully. Instrument agents with tracing to profile behavior, find bottlenecks, and catch quality regressions before customers do. Evals: Build eval sets from production traces, expert examples, and past failures, plus automated checks and LLM judges that measure model and agent quality. Use them to gate every prompt, model, and code change. Feedback & Experimentation: Turn reviewer and customer feedback into measurable improvements through human-in-the-loop workflows, A/B tests, offline replays, and staged rollouts. Technical Ownership: Review designs and code , raise the engineering bar, and contribute to the ML and AI roadmap with Data Science and Platform teams. Required Skills : 5+ years of experience building and running ML or AI systems in production, including at least 1 year on LLM-based applications. Bachelor's or Master's in Computer Science, Maths/Stats, or a related field, or equivalent practical experience. Strong programming skills in Python, including async and concurrent programming, and solid SQL. Hands-on experience building production agents and LLM applications: tool and function design, context window management, RAG and embeddings, structured outputs, multi-step and multi-agent orchestration, and model selection across hosted APIs (OpenAI, Anthropic, Gemini, or similar). Daily use of coding agents such as Claude Code or Cursor, with the judgment to review and verify their output. Solid ML fundamentals (evaluation metrics, overfitting, leakage, drift) and experience deploying models in both batch and real-time setups, including knowing when classical ML beats an LLM. Strong infrastructure skills: Docker, Kubernetes, at least one cloud provider (AWS / GCP / Azure), and production observability (OpenTelemetry, Grafana, Prometheus, or equivalent). Working experience with Terraform and CI/CD, distributed data processing (Spark or Databricks), and workflow orchestration or messaging (Airflow, Temporal, Kafka, or similar). Strong system design skills. You can reason clearly about cost, latency, failure modes, and scale trade-offs. Excellent written and verbal communication, and the ability to drive cross-functional requirements with Product and Engineering. Preferred Skills : Advanced evals : LLM-as-judge calibration against human labels, golden datasets, trace capture and replay, and statistically sound comparisons of variants with confidence intervals. Agent frameworks & protocols: Hands-on with LangGraph, the OpenAI Agents SDK, MCP , or similar, and knowing when a thin custom harness beats a framework. <p data-renderer-start-pos="4411" data-local-id="5