AI Platform Engineer for enterprise AI operations

Duvo Inc; United States; Remote; USD 140,000 - 200,000; Full time

Benefits offered by this job

Remote work

Job summary

Duvo Inc. is building an AI operations platform for enterprise-scale automation, focusing on reliability, cost efficiency, and governance across AI-driven agents.

You will own the AI platform layer, including runtime, context management, tool execution, and observability, ensuring end-to-end production readiness. You will ship systems built on LLMs and retrieval, with ownership of their reliability and performance.

Qualifications

• Production AI systems: scale LLMs in production.
• Prompt engineering, context management, tool use or agent orchestration.
• Own the lifecycle: build, evaluate, deploy, monitor, iterate for AI features.

Responsibilities

• Own the AI platform layer: runtime, context, tool calls, orchestration.
• Ensure reliability, cost and quality in production.
• Design for AI challenges: non-determinism, context windows, failed calls.

Skills

LLM reliability

Prompt engineering

Context management

Tool execution

Agent orchestration

Evaluation pipelines

Observability

Security in AI systems

Open-source tooling

Tools

Haiku models

GPUs

Self-hosted models

Job description

Описание:

A lot of enterprise work still happens by hand. People copy-paste between spreadsheets, live in endless email threads and click through clunky legacy UIs. We started Duvo to end that for good. Customers already trust us with business-critical processes, and our agents are automating them today. We're growing fast, and to win from here we need exceptional people. That's where you come in. We're a team of 60, with 22 engineers, and we just raised a $32 million Series A led by Balderton Capital. What We Are Building Duvo is the AI operations platform for large enterprises. Right now we're focused on retail and consumer packaged goods. Customers build AI agents in Duvo that do the work wherever it lives: SAP, spreadsheets, supplier portals, email, APIs, you name it. We're heavy on browser and computer use. Business users say what outcome they want. The agents plan, act, ask for approval on exceptions and get better with every run. To help customers figure out what to automate, we also map their processes from interviews and internal docs, which gets our foot in the door. We start with the parts of a company we know best (category management, supply chain, finance ops), where we can show value fast, and expand from there to other functions and sectors. Velocity is our moat. We ship fast and iterate even faster.

Задачи:

You'll own the AI platform layer that makes our agents reliable, fast and safe in production. That's the agent runtime, context management, tool execution and orchestration, evaluation pipelines and harnesses, prompt engineering for system behaviors, and the observability tooling that lets our agents do end-to-end work for enterprise customers. You own sandbox behavior and the agent runtime logic. SRE owns the sandbox infrastructure and capacity This is applied AI systems engineering. You ship production systems built on LLMs, retrieval and agent orchestration, and you're accountable for their reliability, cost and quality in the real world. That includes squeezing model usage and cost, like moving work to cheaper models such as Haiku, and running self-hosted open-source models on our own GPUs. AI platform engineers at Duvo are intense about delivering impact fast Our product team is growing into multiple initiatives. Each one has a lead, engineers, a design engineer and an AI-focused engineer.

Требования:

These are the non-negotiables, and it's what we'll evaluate you on: Production AI systems. You've made LLMs reliable at scale in production, beyond research. You've worked hands-on with prompt engineering, context management, tool use or agent orchestration Shipping and ownership. You've taken ambiguous AI problems to production with measurable outcomes. You own the whole lifecycle: build, evaluate, deploy, monitor, iterate System design for AI. You design for what makes AI hard: non-deterministic outputs, context window limits, failing tool calls. Reliability, cost and latency are first-class concerns for you Evaluation design. You build evals that catch regressions, measure quality and give the team the confidence to ship AI features. You understand failure taxonomies and know how to build test sets that mean something Debugging and diagnosis. When things break, you're hypothesis-driven. You trace failures across model behavior, data pipelines and infrastructure until you find the root cause Judgment as AI evolves. You make build-vs-integrate calls on AI infrastructure with incomplete benchmarks, and you course-correct fast as models and providers change You Might Also Have designed and run distributed systems that handle high throughput and complex failure modes Have a strong sense for security in AI systems: prompt injection, insecure output handling, supply chain risks Have contributed to open-source AI tooling or infrastructure projects This Is Not For You If You want an ML research role. We don't train models You mostly want to build user-facing product features and UI. Check out our Product Engineer role instead.

Условия:

EU/UK - Remote A competitive salary and a generous ESOP allocation in a fast-growing company, fresh off our Series A Unlimited spend (within reason) on the AI tooling of your choice If you're remote, we reimburse your co-working space if that's where you feel most comfortable Travel: frequent visits to our Prague office, offsites (our upcoming engineering retreat includes a day at the Porsche Experience Center Hockenheimring) and conferences (we recently sponsored Lisbon AI) Autonomy to do your best work. Want to meet someone to learn from? Set it up. Want a mentor? Go get one A real AI product used by enterprise customers. No demos, no internal tools Many AI modalities in one product: voice agents, computer and browser use, reasoning, and long-running agents. Applied AI, not research A sharp, motivated team that values ownership and candor Trial day (1 day). You get an ambitious problem in the morning, work on it with the team, have a check-in with one of our founders, and present what you built at the end of the day.