ABOUT THE ROLEThis 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.
LOCATIONSingapore or United States. We're hiring in both to reach the right person. Work model is on-site or hybrid, set per location.
CLOSING NOTEIf you don't tick every box but this is clearly your kind of work, get in touch.
