The company
An early-stage, venture-backed software business in New York. It makes security and control software for organisations whose technology and intellectual property carry national security weight, including companies working in AI, defence and other deep tech. The product runs inside customers' day-to-day workflows, so the data it handles is among the most closely guarded these businesses hold.
The job
You'll be the engineer responsible for how the product uses machine learning. You decide which models to use, adapt them to each customer problem and run them on hardware the customer controls. Nothing goes to a hosted model provider. Your systems have to be accurate enough for customers to act on, and every output has to be traceable and explainable.
What you'll work on
- Choosing, adapting and shipping a model for each product use case, and owning that architecture as it grows
- Retrieval pipelines over private company documents and records, built so the data never touches a third-party API
- Fine-tuning open models on customer data where retrieval or prompting falls short
- A set of task-specific agents: some read, label and organise internal documents, others track and analyse outside sources
- Evaluation: test sets, metrics, regression checks and monitoring that show how well each system performs and why
- Self-hosted inference, tuned for latency, throughput and GPU memory on locked-down or offline hardware
- Working with the platform and data engineers so models plug into the product safely
What you'll bring
- At least four years building ML or LLM systems that run in production
- Hands-on fine-tuning of open-weight LLMs (LoRA, QLoRA or full) and a clear method for judging whether it worked
- RAG built end to end: retrieval design, chunking, choice of embeddings and measuring answer quality
- Multi-agent systems in production, including tool calling, orchestration, failure handling and tracing
- Experience serving models yourself (for example with vLLM, TGI or llama.cpp) where calling an external API isn't an option
- Solid software engineering: tested, maintainable code other engineers can build on
- A bonus: document classification, entity extraction or NLP at high volume
What's on offer
- $200k to $295k base depending on experience, plus equity at an early stage
- Medical, dental and vision insurance
- Working directly with the founder, with a real say in where the product goes
- A small team, so you shape how machine learning is done here
Location and eligibility
New York City, in the office most of the week. If you're elsewhere in the US, you can start remotely and move later, with help towards relocation. The work involves export-controlled technology, so you'll need to be a US citizen or green card holder, or otherwise qualify as a US person.
If this sounds like your kind of problem, send me a message and I'll share more.
