Software Engineer, AI Architecture

About the role

A model on its own is not a product. What turns it into one is the architecture around it: how work is broken into steps, which tools an agent can reach, what context it sees, how its output is checked and what happens when it fails. As a Software Engineer in AI Architecture, you will design and build those systems for Zocri and for the businesses we work with.

Zocri runs agents that make real apps and websites in isolated sandboxes, routes work across several model providers and checks what they produce before a person sees it. You will make that faster, cheaper and more reliable, and help our clients build systems of the same quality inside their own companies.

What you will do

• Design and build agent systems: planning, tool use, memory, sandboxed execution and the hand offs between them
• Build model routing and fallback across providers, balancing quality, latency and cost for each kind of task
• Design retrieval and context pipelines so agents see the right information at the right time
• Create evaluation frameworks and automated checks that measure whether an AI system does its job, and run them on every change
• Make AI systems observable with tracing, cost and latency metrics, and the tools to understand why a run went wrong
• Lead architecture decisions and reviews on client projects, and turn what works into patterns the whole company reuses

You may be a good fit if you

• Have 5+ years of software engineering experience building and running production systems, including at least 2 years building with large language models
• Have shipped agentic or LLM powered features that real people use, and have opinions formed by what broke
• Are strong in TypeScript or Python, and comfortable with distributed systems, queues, caching and streaming
• Understand the trade offs between prompting, retrieval, fine-tuning and tool use, and know when each is the right answer
• Treat evaluation as engineering: you write tests for model behaviour, not only for code
• Can explain a complex system clearly to engineers, product managers and clients

Strong candidates may also have experience with

• The Model Context Protocol, tool calling and structured outputs across several model providers
• Running untrusted or generated code safely in sandboxes or containers
• Security for AI systems, such as prompt injection defence and secret handling
• Technical leadership on client facing projects

About NerfedLab

NerfedLab builds AI products, helps businesses put AI to work in their everyday operations, and runs a research team that publishes its work in the open. Zocri, our product, lets anyone make an app or a website by describing it, then run the business behind it: sign-in, payments, messages and marketing in one place.