Full-time

Internal AI Architect

About Ravical

At Ravical, we're transforming how professional services firms (accountancy, tax, audit, insurance, payroll) deliver and scale their work, by building AI-first, agentic products. We sell work, not software: our product exists because we build it inside our customers' operations across Europe. We're a small, deliberately dense team (~30 people) scaling from pre-seed toward Series A, with a culture built on five things that don't change: talent density is the strategy, AI-native rather than AI-assisted, clarity over comfort, outcomes instead of activities, and built with customers, not for them.

If you're excited by hard problems, practical AI, and visible impact from week one, we'd love to meet you.

What you’ll do

• Ship the MVPs. Turn what teams need into capabilities that live in the shared system, and get handed off so the team runs them without you.

• Evals & quality. Own the discipline of measuring whether what we build is actually good, and own the switch to new models and tooling as the frontier moves (always the right model for the job, not the most expensive).

• The shared integration & data layer. Where Notion, HubSpot, Fathom, Outlook, Linear and the rest meet. Build the connectors that don't yet exist, unblock getting data out of our own systems, and put the right access in the right hands.

• Output governance. Decide what reaches Slack versus stays agent-to-agent. Cut noise, make outputs actionable, get the right information to the right person in the right place.

• Security review. A documented, repeatable gate baked into how agents and connections ship.

• Adoption. Be metric-driven about it: is it used? If not, why, and fix that.

Who you are

• A strong, pragmatic engineer who can own a system end-to-end, with as much product-manager instinct (reads cross-team processes, prioritises, tracks adoption, clears blockers) as engineering.

• A real platform engineer: you can ship production software to a team's standard, not only standalone internal agents.

• Hands-on with LLMs and agentic systems: you've built a production-grade, multi-step agent that other people depend on, with live connections, a permissioning/security boundary you designed, and proof someone other than you relies on it.

• Disciplined about evals: you can point to how you measured whether an agent was good, and a model or tooling switch you made on the back of it.

• Someone who distributes leverage rather than becoming the bottleneck. Your best work keeps running after you've moved on.

• Mid-to-senior. Strong Python or TypeScript skills.