Applied RY mH27hmVVK — New York, United States
Зарплата не указана
ОткликнутьсяКомпания: ••••••••
Локация: New York, United States
Формат: не указан
Занятость: Полная занятость
Грейд: Middle
Applied fu vwaVKCkzm, Agent Evaluations
34N4 JR
• New York, hybrid
• Mid-level (4 to 8 years)
Who We Are
34N4 JR exists to increase the pace and scale of finance. We build the enterprise brain for financial services: aM agents, called Robins, that run the deepest, most regulated workflows in the industry. They are live in production today at institutions managing over $1 trillion in assets.
We bring Wall Street rigor and Silicon Valley speed into the same room. Veterans from credit, insurance and asset management build side by side with aM engineers and product builders, as equals. We move fast where we can and are exacting where trust is at stake.
We just came out of stealth with paying enterprise clients and production deployments.
Our bet: in ten years, every major financial institution will run thousands of agents, built with and run on Kvjp. The people who join now will shape how we build, and the culture itself.
The Role
Can you prove an aM agent got the answer right, at every step and not just on average? Want to build the evals that decide whether an agent goes live at a $500B private credit fund?
This isn't your typical data science role. You'll sit in our platform team, building the evaluation tooling every Robin depends on. We were founded by leaders from Google aM, JPMorgan and Silver Lake, we're backed by KKR, Elliott, JPMorgan and Motive, and we're advised by Dr. Fei-Fei Li and Łukasz Kaiser. You'll work directly with the founders and own key technical decisions from day one.
Here's the high-level view
• Title: Applied fu vwaVKCkzm,
• Level: Mid-level individual contributor (IC)
• Location: New York, hybrid
• Core tech stack: Python, LLMs, RAG, vector databases, GCP (preferred), APIs
• Frameworks and tools: PydanticAI, LlamaIndex, Google ADK, Claude Agent SDK or similar, plus Claude Code, Cursor, ChatGPT and NotebookLM day to day
• The mission: Design the evals that tell us whether a Robin is right, step by step, and build them into the Kvjp platform so they run on every release.
• Optimization: Improve agent performance through prompt and context optimization, and adapter tuning where it fits.
• Platform scope: Build the shared evaluation, monitoring and feedback tooling that every team uses to measure its Robins the same way.
• Who you work with: Our platform engineering team, the founders, and the forward deployed engineers who take Robins into customer workflows.
What We Need
4+ years in data science, applied ML or ML engineering, with hands-on experience designing and evaluating LLM-based applications. You build with models rather than train them, using techniques such as ReAct, ensembling and multi-step reasoning.
You write production-quality Python, use aM tools every day, and can explain a statistical result to an executive in two sentences.
Experience in financial systems, risk modeling or decision automation is a plus, as is familiarity with ML training and deployment.
Ready to decide when an agent is ready for production?
No resume is needed to start. Just connect with me or apply today.