> Markdown version of [/jobs/ext/4618125-staff-machine-learning-engineer](https://www.wearedevelopers.com/jobs/ext/4618125-staff-machine-learning-engineer). Every page supports `.md` or `Accept: text/markdown`. Links point to the HTML versions so they work for humans too. Agent guide: [/agents.md](https://www.wearedevelopers.com/agents.md).

---

# Staff Machine Learning Engineer

- **Company:** 53 Stations
- **Location:** Boston, MA, United States
- **Experience:** Expert
- **Contract:** Permanent contract
- **Skills:** LangGraph Framework, Application Programming Interfaces (APIs), Artificial Intelligence, Amazon Web Services, Audit Trail, Microsoft Azure, Data Stores, Cursor, Information Retrieval, Python (Programming Language), PostgreSQL, Machine Learning, Language Modeling, NumPy, Swagger, Search Technologies, Openapi, Pytorch, Large Language Models, Claude Code, Multi-Agent Systems, Apache Spark, Generative AI, AI Coding Agents, Git, Pandas, LangSmith, Scikit Learn, Kubernetes, Information Technology, HuggingFace, Machine Learning Operations, FAISS, VLLM, OpenSearch, Docker, Databricks
- **Published:** November 7, 2026
- **Apply:** https://www.juju.com/job/28_25197_6c792ae538895852ff85a1df92bf8be6

## About the Role

* BS/MS in Computer Science, Machine Learning, or a related field, or equivalent experience
 * 8+ years building and shipping ML systems in production, with a proven track record of delivering product-oriented ML solutions, including recent work on LLM-powered products
 * Deep expertise in PyTorch and the modern Python ML stack (Hugging Face, scikit-learn, NumPy, Pandas/Polars)
 * Hands-on experience building production LLM applications: retrieval-augmented generation (RAG), tool/function calling, structured outputs, and agentic workflows
 * You've shipped agentic systems in production and know their failure modes: tool use, planning, state, recovery and human-in-the-loop
 * Experience post-training or adapting LLMs (SFT, DPO, RLHF/RLAIF, reward modeling) for a production use case
 * Rigorous about evaluation: you've built eval suites, golden datasets and LLM-as-judge pipelines
 * Sound judgment on when to use a frontier model API, a fine-tuned open-weight model, or a classical ML model, balancing cost, latency, and accuracy
 * Passionate about applying SOTA techniques to real-world problems, with a willingness to get your hands dirty in the data and domain to achieve measurable outcomes
 * You care deeply about engineering excellence, supportability and knowledge-sharing
 * You have strong written and verbal communication skills
 * Heavy user of AI coding tools (Cursor, Claude Code) in your daily workflow

Nice To Have, But Not Required

 * Building multi-agent systems and agent tooling (e.g., LangGraph)
 * Hybrid retrieval: vector search, re-rankers, and embedding fine-tuning
 * Vision-language models for document understanding (invoices, bills of lading, customs forms)
 * Multilingual NLP
 * LLMOps: MLflow, Databricks Mosaic AI, and guardrails
 * Experience with safety, guardrails or auditability for AI in regulated or high-stakes domains (e.g., customs, legal, healthcare, finance)

Technologies we love

 * Languages: Python, Spark
 * Tools: Docker, Git, Kubernetes, Swagger/OpenAPI, AWS, Azure
 * Datastores: Databricks, OpenSearch, Postgres
 * ML Tools: PyTorch, Hugging Face, FAISS, MLflow
 * LLM Tools: vLLM, LangGraph, LangSmith/Braintrust, frontier model APIs

This role is remote or hybrid, based in the United States, UK, with the option to work from one of our hubs in New York City, Boston, Washington D.C., or London.

## Description

* Drive the technical direction for agentic systems. Define the architecture and roadmap for the models and agentic systems that process trade data as it arrives (shipments, customs declarations, invoices, bills of lading) and turn it into risk, classification and compliance decisions in real time.
 * Build the pipeline from ingestion to decision. Lead the design of reliable, cost-efficient systems for ingestion, document understanding, entity resolution, inference and routing across models, at the scale of global trade.
 * Adapt models to trade. Use analyst feedback, production outcomes and agent traces to fine-tune and post-train models (SFT, DPO, RL with domain graders) where it measurably beats prompting, and keep every model improving over time.
 * Design the observability and feedback stack. Build the evals, tracing and feedback loops that decide what ships and catch regressions in production.
 * Lead beyond your team. Mentor senior engineers, raise the technical bar across Engineering, and help shape how we hire and level ML talent.
 * Shape the product and the business. Work with software engineers, data scientists, product managers and customers to generate new business ideas and turn ambiguous problems into shipped capabilities.