Senior AI Engineer, Agent Capabilities

Our agents hold real conversations with the people running fraud at scale. You own the harness around those agents: what they can do, how well they hold a conversation, and how human they feel to the person on the other side. Today they work in text. You will take them to voice, to live calls and to email, and make an agent able to change channel mid-conversation without losing the thread.

About CUBE AI

CUBE AI is an Agentic Crime Prevention Platform, built at the intersection of agentic AI, cybersecurity, and fraud prevention. We serve the institutions on the front line of an exploding wave of AI-powered crime: banks, fintechs, and the platforms criminals cannot operate without. Our mission is ambitious: make the world a safer place. How we do it is the part we're most excited about, and the part we're happy to walk you through in a conversation rather than on the open internet.

Why this is the seat to take

This is a rare seat on a rocket ship. CUBE AI is early, exceptionally ambitious, and backed by some of the world's leading investors. Our founders have built and scaled startups to unicorn valuations and exits, and have run mega teams at mega companies. We are building the company that puts the institutions the modern economy runs on back on offense against AI-powered crime. The closest analogy is joining Palantir in its early days: a hard mission, a category we're defining rather than entering, and a small group of exceptional people compounding fast. If you want a comfortable role, this isn't it. If you want the defining chapter of your career, your best work alongside people who make you sharper, keep reading.

What you'll do

• Own conversation quality, and be accountable for how human our agents feel in a real exchange.
• Take agents from text to voice, including live calls.
• Make an agent switch channel mid-conversation without dropping the thread or the context.
• Add new channels, and make adding the next one cheaper than the last.
• Turn recorded conversations into evaluation sets and training data that measurably improve the agents.
• Build the evaluation harness that proves a change made the agents better rather than merely different.
• Partner with our models engineer, who comes at the same target from the model side.

Our stack

• Python, LangChain and LangGraph
• Frontier LLM APIs, with structured output through Pydantic
• FastAPI services, with PostgreSQL-backed conversation state
• GKE, Docker and GitHub Actions
• Realtime voice is the part you will build. There is no incumbent stack to inherit.
• AI-assisted, agentic development workflows

What we're looking for

The bar (non-negotiable, every role):

• Extreme drive and passion for the work and the mission.
• Extreme ability to learn fast and operate with confidence in uncertainty.
• Extreme IQ and EQ. You raise the level of every room, and you lose an argument gracefully the moment someone's right.

For this role specifically:

• You have built LLM-based systems in production rather than in notebooks.
• Demonstrable taste in conversation design: you can say why a bot sounds like a bot, and then fix it.
• Experience with realtime or streaming audio, or the appetite to own it from nothing.
• Evaluation discipline. You do not ship a prompt change without a way to know it helped.
• You know modern cost and context optimization techniques to have the pipeline not only effective but also efficient
• Strong Python.
• Fluent written and spoken English.

Bonus points

• Speech-to-text and text-to-speech pipelines, and latency work on them.
• Telephony: SIP, WebRTC or a provider API.
• Multi-agent or tool-using architectures.
• An adversarial domain: fraud, security or abuse.

How we work and who thrives here

Our core locations are Lithuania, London, and New York, and we consider remote-first candidates who can travel as needed for their role. We write more than we meet and move at warp speed. We're an AI-first company: every person here is amplified by a layer of AI agents, and we scale capability, not headcount. We don't buy services, we buy results. We're drivers, not passengers, so we hire people who go and do it without waiting for permission. We move fast, and we use judgment where it counts: we hold sensitive data and the trust of the institutions we protect, and we never cut corners there. Our culture runs on five virtues:

• A pedantic bar for quality. Work that is thorough, complete, and on time. We trust each other to deliver, fully.
• Radical ownership. See it, own it, fix it. Never defer a problem or an opportunity. You will never be reprimanded for taking initiative.
• Relentless energy and audacity. Building something exceptional demands enormous energy. We move forward with maximum drive.
• Automate everything. We scale the capability of our people, not the headcount. We always hunt the 10x multiplier, working alongside AI agents rather than piling on process.
• Have fun. We work with people we genuinely enjoy. It gets us through the hard stretches and keeps politics out.

What's in it for you

• Uncapped growth. In a company moving this fast, scope finds the people who can carry it. Ownership and title are earned in months, not years.
• Real self-realization. This is a place to become the most capable version of yourself, next to people operating at the top of their game.
• A direct line to senior product and engineering leaders, and a real say in where the product goes.
• You share in the outcome. When the company wins big, you win big.

What we offer

Join a rocket ship. Let's discuss compensation and the rest of the details on our call.

CUBE AI is an equal-opportunity employer. We welcome applicants of every background and will support any accommodations you need during the hiring process.