Job description
Description
This is an Atlantic Labs portfolio venture. Find out more about us and other opportunities in our portfolio here.
About the Venture
We’re a stealth venture building an AI-native platform for enterprise business modelling and scenario planning. At its core is a graph-based modelling engine combined with an intuitive grid UI and an AI co-pilot that accelerates model creation and improves decision quality.
Our goal is to replace Excel as the default for complex business models, while offering a more flexible and easier alternative to current heavyweight enterprise tools. It’s a new kind of tool for business controllers and FP&A leaders.
We’re a founding team of three with deep expertise in enterprise software, data modelling, and product execution. Backed by pre-seed financing, we’re working closely with design partners in the upper mid-market and enterprise space, and preparing to launch soon.
We’re establishing our HQ in Helsinki and would be excited to hear from you who are already based here or open to relocating. That said, we are open to hybrid arrangements within a reasonable aligned timezone.
About the Role
Weʼre looking for a hands-on AI/LLM engineer to own the design and implementation of the platformʼs AI capabilities end to end. Youʼll be responsible for how models reason over our grid + graph, for safety and reliability in enterprise environments, and for building agentic workflows that deliver practical value in complex modelling and variance analysis.
This is a zero-to-one role where youʼll pair research-grade curiosity with product pragmatism. You will collaborate closely with the founding team, shipping iteratively while laying the foundations for correctness, observability, and scale.
What You’ll Do
Design and implement agentic workflows for model building, scenario generation, reconciliation and variance analysis
Build reliable, auditable LLM pipelines: retrieval, tool-use, function calling, routing, and self-correction loops
Integrate LLMs with our graph-based modelling engine and domain-specific DSLs and tools
Establish evaluation frameworks: unit-style evals, golden sets, regression tests, and human-in-the-loop review
Productionize prompt and policy governance for enterprise: safety, privacy, red-teaming, guardrails
Own LLMOps: tracing, observability, cost and latency optimization, caching, batch inference
Ship high-impact product features in tight loops with design partners and founders
Contribute to core architecture decisions balancing speed with reliability and compliance
Requirements
About You
Youʼve implemented LLM-powered features in production enterprise applications
Hands-on experience building agents or complex tools - using systems beyond basic chat
Strong software engineering fundamentals and product sense; you ship, measure, iterate
Comfortable moving across the stack when needed and collaborating cross-functionally
Pragmatic about model selection and trade-offs; you value reliability over hype
Nice to have
Experience with graph-based data models, planning algorithms, or numerical modelling
Retrieval design at scale, vector indexing strategies, hybrid search, and metadata filtering
Practical grounding in evals and offline experimentation methods
Prior work in FP&A, analytics, supply chain, or other multi-SKU enterprise domains
What success looks like in 3–6 months
Agentic workflows reliably generate and explain scenarios over our graph + grid foundation
Evals and traces catch regressions early; we can ship AI changes with confidence
Enterprise-grade safeguards are in place: policy checks and auditability
Measurable improvements to model creation time and decision quality for design partners
Why Join?
This is a true founding opportunity. Youʼll shape the AI architecture, agent patterns, and evaluation culture from day one, working on problems that are technically hard and meaningful for customers. If you want to define how enterprises reason about decisions with AI — and build a system that replaces spreadsheets for the real world — weʼd love to talk.
Location
Helsinki strongly preferred. Remote or hybrid possible within close timezones.
