Senior/Staff Applied AI Engineer

Storm2; New York (NY); In loco; USD 250.000 - 320.000; Tempo pieno

Vantaggi offerti da questo lavoro

Equity

Descrizione del lavoro

Storm2 in New York City is seeking a Staff Applied AI Engineer to own model selection, fine-tuning open-weight LLMs, and building production agents with guardrails.

You’ll stay hands-on in Python and PyTorch, evaluate models for accuracy, latency and cost, and guide the technical direction for applied AI across the business. This role is in-office in NYC with equity on offer.

Competenze

• Proven experience fine-tuning LLMs and shipping agents to production.
• Strong Python and PyTorch skills, with experience in Hugging Face.
• Ability to balance model quality, latency, and cost.
• Experience with open-weight models and evaluation at scale.

Mansioni

• Fine-tune open-weight LLMs with LoRA/QLoRA, SFT and DPO.
• Build production agents with tool usage and guardrails.
• Convert investigator feedback and agent traces into training data.
• Benchmark models for accuracy, latency, cost and hallucinations.
• Deploy and optimize models using vLLM, TGI, Triton or similar.

Conoscenze

Fine-tuning LLMs

Python

PyTorch

Hugging Face

Production agents

Model evaluation

Strumenti

LoRA

QLoRA

SFT

DPO

vLLM

TGI

Triton

Guardrails

Quantisation

Distillation

Graph ML

Descrizione del lavoro

Staff Applied AI Engineer – LLMs & Agents

New York City, NY (In-office)

$250k–$320k Base + Equity

The Company

Storm2’s client is a claims intelligence startup using AI and data to uncover fraud across property and casualty insurance. They’re solving a difficult problem where the data is messy, fraudsters adapt, and false positives have real consequences.

The Role

Frontier models work, but they’re expensive and slow at scale. Smaller open-weight models solve that problem, but only if you can preserve the reasoning quality.

That trade-off is the job.

You’ll own which models the company builds on, fine-tune open-weight LLMs for specific workflows, put them inside production agents, and build the evals that decide what actually ships. This is a Staff IC role with genuine architectural ownership, while staying hands‑on in Python and PyTorch.

What you’ll be working on:

• Fine-tuning open-weight LLMs with LoRA, QLoRA, SFT and DPO.


• Building production agents with tool use, structured outputs and guardrails.


• Turning real investigator feedback and agent traces into training data.


• Benchmarking models on accuracy, hallucinations, false positives, latency and cost.


• Serving and optimising models through vLLM, TGI, Triton or similar.


• Setting the technical direction for applied AI across the business.

You’ll be good here if:

• You’ve actually fine-tuned LLMs and can explain what moved the numbers.


• You’ve shipped agents into production, not just built demos.


• You’re strong in Python, PyTorch and the Hugging Face ecosystem.


• You’ve made real trade-offs between model quality, latency and cost.


• You enjoy problems where the ground truth is messy and the answer isn’t sitting neatly in a research paper.


• Experience with quantisation, distillation, entity resolution, graph ML or highly imbalanced detection is a big plus.

Why apply?

You won’t be building another wrapper around a frontier API. You’ll decide which models run which workflows, how they’re trained, and whether they’re good enough to replace much larger models at production scale. At $250k–$320k base + equity, they’re paying for someone who can make those calls and then build the systems themselves.