Senior Applied AI Engineer – Agent Runtime (Alwin by DataSnipper)

We are looking for a Senior Applied AI Engineer to join the Agent Runtime team behind Alwin, our new Agentic Automation Platform for audit and finance. Alwin agents run for long times, work through multi-step audit procedures over client evidence, and return finished work papers with every number traced back to source. Humans review and sign off.

The runtime is the layer every agent depends on: the model access, the base system prompt and context, the tools (document extraction, retrieval, sandboxes, MCP integrations), the orchestration of agents and sub-agents, and the observability and evals that tell us whether an agent is performing to objective standards. You will own the applied AI half of that layer. Product teams build audit-specific agents on top of it; you decide how the base agent reasons, what tools it gets and at what abstraction, how it manages context over long horizons, and how we measure and hill-climb accuracy, speed and cost.

This is a hands-on role in a small team with a large blast radius. Your work goes in front of hundreds of thousands of audit and finance professionals, and the problems are largely open: there is no playbook for production agents in a regulated domain, so you will help write it.

ABOUT DATASNIPPER

DataSnipper is the Agentic Automation Platform for audit and finance. Known worldwide for our Excel add-in, we are now building Alwin by DataSnipper: purpose-built agents that execute audit and finance workflows end-to-end, with every output traceable back to source evidence and a human signing off. Headquartered in Amsterdam with offices in New York, Tokyo, Kuala Lumpur and Sydney, we are used by hundreds of thousands of professionals at the world's largest firms. Our mission: automate the mundane, unlock the meaningful.

WHAT YOU WILL DO

AGENT ENGINEERING

- Own the base agents: system prompt, context engineering, memory and state management for tasks that span many turns and hours of execution

- Design how agents use the tool layer, including document extraction, retrieval, code and file sandboxes, and MCP integrations, choosing the right level of abstraction so agents handle edge cases without wasting effort on mechanical steps

- Implement agentic patterns such as sub-agent composition, planning modes, human-in-the-loop gates, and model routing across multiple providers

- Ship end-to-end: from prototype with our audit domain experts, through evaluation, to production on Alwin

- Define and build automatic processes that improve the agent continuously over time

- Keep costs under control and implement cost-efficient approaches to AI-centric workflows

EVALUATION & QUALITY

- Extract signal from long agent trajectories: attribute outcomes to specific reasoning steps and tool calls, classify failure modes, and turn them into fixes

- Hill-climb accuracy, latency and token cost, and make the trade-offs explicit for the teams building on the runtime

RELIABILITY & COLLABORATION

- Instrument agent behaviour with our observability stack and use production traces to drive improvements

- Partner with Agent Experience teams, Document Intelligence and the AI Platform team to turn their needs into runtime capabilities that are self-service rather than a request queue

- Stay on the frontier: evaluate new models, techniques and agent patterns, and bring the ones that hold up into production

WHAT YOU WILL BRING

MUST-HAVE

- 5+ years of software engineering experience with strong, production-grade Python

- Experience shipping and operating an LLM-powered product in production: you have dealt with hallucinations, latency spikes, tool failures and cost explosions at scale, and can explain what broke and how you fixed it

- Hands-on experience building agentic systems: control loops, tool selection, planning versus execution, retries and fallbacks, not only prompt-and-parse pipelines

- Experience with evaluation: you have built datasets, run offline and online experiments, and u …