Data Scientist Intern, AI Agents Filled Internship

Arctal builds structured datasets from unstructured financial documents—100,000+ PDFs (fund reports, regulatory filings, investor letters) turned into clean, queryable data that institutional buyers use for decision-making.

AI agents do the reading. We build the agents. Team of 5, output of 50.

A dataset is not a fact. It's a representation of reality that someone chose to stand behind. AI agents do the extraction—humans make the judgement calls. You'll learn what that means in practice.

Our customers are asset managers, banks, and financial data firms who need reliable data extracted from documents that were never meant to be machine-readable.

You’ll work with the Data Scientist and technical team to deliver datasets to clients. This is your introduction to the super-IC model—learning what it means to direct agents rather than do the work yourself.

Data Quality. Validate data outputs, catch errors, and flag issues before data goes to clients. You're the human-in-the-loop that makes sure quality is high. You'll develop an eye for what great data looks like—and what's subtly wrong.

Agent-Led Data Work. Use Claude Code, Cursor, and other AI tools to process documents, clean data, and build small automations. You'll spend a lot of time in the terminal, in Python, and in messy datasets—and that's the point. The technical team will help you level up.

Process Improvement. Spot patterns in what's slow or broken. Suggest fixes. If you find a better way to do something, we'll use it. At Arctal, every person is building themselves out of their current role—automating the task they did yesterday so they can take on the harder problem tomorrow.

This is a hands‑on role. You'll have real responsibility from day one—not coffee runs and slide decks.

Final-year student, master's student, or early-career—what matters is what you can do, not your CV.

Data-obsessed. You've spent real time working with data—cleaning it, analyzing it, understanding why it's wrong. You know the difference between data that looks right and data that is right. You notice when numbers don't add up.

Technically deep. You're fluent in Python and comfortable in the terminal. You've written scripts, built pipelines, or wrangled messy datasets. You don't need hand-holding on technical work.

AI-native (for real). You actively use AI coding tools—Cursor, Claude Code, Windsurf, Cline, or similar. You know the difference between chat-based AI and agentic coding. Your first instinct facing a new problem is to design a system that solves it autonomously. You've probably broken something by letting Claude run too many commands unsupervised.

What we're filtering for. We need someone who is technical and data-first. Previous hires that didn't work out were people who weren't deep enough in data or weren't fluent with AI coding tools. If you haven't spent significant time in the terminal, in codebases, and in messy datasets, this isn't the right role.

What matters: curiosity, attention to detail, bias to action. You care about getting things right, not just getting things done. You see AI agents as leverage, not a threat—and you're excited to be the one building them, not the one being replaced by them.

• An assistant role (you’ll own real deliverables)
• A learning-only experience (you’ll ship to real clients)
• A 9-to-5 (intensity is high, learning is faster)

Aleksi (CEO) — Cambridge engineering + ML. Co-founded Secondmind, founding team viously worked on ML with Carl Rasmussen at Cambridge.

Krista (CCO) — Former Head of Market Intelligence at Climate Bonds Initiative. Deep sustainable finance and capital markets expertise.

Small team that ships fast. No layers, no politics—just building.

• Old Street office.
• Real work with real clients from week one.
• Freedom to use AI tools however you want.
• A team that will teach you how startups actually work.

For strong performers, this converts to a full-time Data Scientist role managing your own fleet of agents.