Job description

ABOUT ELEMYNT ELEMYNT is an early-stage startup built by Xora Innovation. We develop applied intelligence that brings AI into the real world. Our platform combines advanced machine learning, high-performance simulation, and modern software engineering to accelerate the design, validation, and deployment of new materials. Our work sits at the intersection of AI, physics, and large-scale computation. The problems are hard, the stakes are high, and the impact is tangible. ABOUT THE ROLE This role owns the LLM systems behind our platform: the agents and fine-tuned models that ship as product, and the engineering that keeps them reliable — evaluation, tracing, and production-quality services. It's deeply hands-on, from model internals to shipped software. The platform runs inside our customers' own secure environments: their compute, their cloud, or a hybrid. So the LLM layer has to work with commercial APIs and self-hosted models alike, and carry its own safeguards wherever it lands. Every LLM capability we ship stands on this work. WHAT YOU WILL DO • Build and ship LLM-powered capabilities end to end: prototype, evaluate, deploy, and iterate them into production services users rely on. • Design agents that plan and carry out multi-step work: tool calling, structured outputs, durable state, and the judgment to know when an agent is the wrong tool. • Build retrieval that gives models the right context: ingestion, chunking, embeddings, hybrid search, reranking. • Fine-tune open-weight models with LoRA, QLoRA, or full-parameter tuning on multi-GPU, curating the training data and choosing the method by task, compute budget, and target. • Build evaluation loops that gate what ships: automated scoring, LLM-as-judge, and regression tracking against curated test sets. • Instrument model calls and tool use with tracing, so quality, cost, and failures stay debuggable in production. • Turn LLM capabilities into clean APIs and reusable tooling that other engineers build on. WHAT WE ARE LOOKING FOR • Bachelor's or Master's degree in Computer Science or a related engineering field, and 5+ years building and shipping production software, including deep hands-on work building LLM-powered systems in production. • Strong Python and a track record of shipping reliable services: async, HTTP and streaming APIs, testing, code review. • Production experience with LLMs: prompting and context engineering, tool calling, structured output, and the latency and cost work that keeps them usable. • Hands-on experience designing and shipping agents: the loop, the tools, context, memory, and where they fail. A framework such as LangGraph or equivalent; structured outputs in Pydantic or JSON Schema. • Experience building RAG systems: embeddings, chunking, hybrid search, reranking, and a feel for what actually moves retrieval quality. • Direct experience fine-tuning open-weight models (LoRA, QLoRA, or full-parameter) on multi-GPU, including curating and formatting the training data. • Experience with LLM evaluation and guardrails: LLM-as-judge or automated scoring, regression tracking, and tracing over agent runs. • Experience building shared LLM tooling or platform components that other engineers build on, and comfort owning ambiguous systems end to end in an early-stage environment. NICE TO HAVE • Self-hosted inference with vLLM, TGI, or SGLang, served behind an OpenAI-compatible interface. • Interoperability standards for tools and agents, such as MCP. • Retrieval over structured data: knowledge graphs, hybrid search, reranking at scale. • LLMs applied to scientific or other technical data; experience making APIs and tool surfaces easy for agents to call reliably. • Contributions to open-source AI/ML: agent frameworks, eval tooling, RAG, fine-tuned models. LOCATION Singapore or United States. We're hiring in both to reach the right person. Work model is on-site or hybrid, set per location. CLOSING NOTE If you don't tick every box but this is clearly your kind of work, get in touch.

Requirements

Must-haves

Build and ship LLM-powered capabilities end to end: prototype, evaluate, deploy, and iterate them into production services users rely on.
Design agents that plan and carry out multi-step work: tool calling, structured outputs, durable state, and the judgment to know when an agent is the wrong tool.
Build retrieval that gives models the right context: ingestion, chunking, embeddings, hybrid search, reranking.
Fine-tune open-weight models with LoRA, QLoRA, or full-parameter tuning on multi-GPU, curating the training data and choosing the method by task, compute budget, and target.
Build evaluation loops that gate what ships: automated scoring, LLM-as-judge, and regression tracking against curated test sets.
Instrument model calls and tool use with tracing, so quality, cost, and failures stay debuggable in production.
Turn LLM capabilities into clean APIs and reusable tooling that other engineers build on.
Bachelor's or Master's degree in Computer Science or a related engineering field, and 5+ years building and shipping production software, including deep hands-on work building LLM-powered systems in production.
Strong Python and a track record of shipping reliable services: async, HTTP and streaming APIs, testing, code review.
Production experience with LLMs: prompting and context engineering, tool calling, structured output, and the latency and cost work that keeps them usable.
Hands-on experience designing and shipping agents: the loop, the tools, context, memory, and where they fail. A framework such as LangGraph or equivalent; structured outputs in Pydantic or JSON Schema.
Experience building RAG systems: embeddings, chunking, hybrid search, reranking, and a feel for what actually moves retrieval quality.
Direct experience fine-tuning open-weight models (LoRA, QLoRA, or full-parameter) on multi-GPU, including curating and formatting the training data.
Experience with LLM evaluation and guardrails: LLM-as-judge or automated scoring, regression tracking, and tracing over agent runs.
Experience building shared LLM tooling or platform components that other engineers build on, and comfort owning ambiguous systems end to end in an early-stage environment.
Singapore or United States. We're hiring in both to reach the right person. Work model is on-site or hybrid, set per location.
If you don't tick every box but this is clearly your kind of work, get in touch.

Nice-to-haves

Self-hosted inference with vLLM, TGI, or SGLang, served behind an OpenAI-compatible interface.
Interoperability standards for tools and agents, such as MCP.
Retrieval over structured data: knowledge graphs, hybrid search, reranking at scale.
LLMs applied to scientific or other technical data; experience making APIs and tool surfaces easy for agents to call reliably.
Contributions to open-source AI/ML: agent frameworks, eval tooling, RAG, fine-tuned models.

Application prep

Use the role’s language and application details while tailoring your materials.

Keywords to mirror

aimachine learningllmpythonml

Platform: workable
Effort: medium