LLM Engineer (Visa Sponsorship)

New York Cityfull timeOn Site

Apply on Employer Site

H1B: yes

Spherecast LLM Engineer at Spherecast builds Agnes, an AI Supply Chain Manager, owning AI systems end-to-end from prototypes to production. The role involves designing LLM-powered automations and evaluation pipelines for supply chain optimization.

Responsibilities

Build Agnes AI Supply Chain Manager from the ground up
Own AI systems end-to-end from prototypes to production
Design and maintain evaluation pipelines for prompts and workflows
Build LLM-powered automations or agents core to production systems
Implement LLM observability and monitoring to improve accuracy

Qualification

You are a great fit if you haveBonusAbout SpherecastSpherecast is building AgnesAgnes decides what to produce

Required

You are a great fit if you have:
Hands-on experience with modern LLM APIs (e.g. Anthropic, OpenAI, DeepSeek, OpenRouter, Gemini, Moonshot) and have shipped features using them.
Strong intuition for large language model selection – you understand the strengths, weaknesses, latency/cost tradeoffs, and ideal use cases of different LLMs.
Practical experience with the HuggingFace ecosystem in real projects.
A track record of building LLM-powered automations or agents that are core to a production system, not just internal demos or playgrounds.
Experience designing and maintaining evaluation pipelines to iterate quickly and safely on prompts and workflows.
Strong prompting and system-design skills – you know how to design tools/functions, structured outputs, and multi-step agent flows that are robust to edge cases.
Experience with LLM observability and monitoring (logging, traces, quality metrics, feedback loops) to track and improve production accuracy over time.
Bonus: Experience running self-hosted LLMs in production or serious prototypes.
About Spherecast
Spherecast is building Agnes, the AI Supply Chain Manager for Consumer Goods.
Agnes decides what to produce, where to make it, and how to move it through factories, warehouses, and channels – simulating thousands of options in seconds and executing the best plan.