Tasks

What you'd work on

Two hard problems, both in production.

AI-native project management, not a chatbot on PDFs. You build the product project managers live in every day: the dashboards, task views, and alerts. The AI belongs in the background, not in a sidebar you talk to, so you're building surfaces where automation runs on cronjobs, recommendations surface for the PM to approve, and some tasks run end to end with the user only reviewing after.

The data layer under a five-year project. Construction data is deeply interconnected: documents, protocols, decisions, tasks, and stakeholders all cross-reference each other across timelines that run for years. You design the schema, write the queries, and build the APIs that hold that graph and still serve the product fast enough to feel instant, at hundreds of thousands of pages and 50 stakeholders per project.

Stack. TypeScript, Next.js, Vercel, Supabase (Postgres + pgvector), LangChain, Vercel AI SDK, LangFuse, shadcn/ui. Every engineer gets €500/month for AI tooling: Claude Code, Cursor background agents, and whatever frontier model you want to test. No legacy. Greenfield.

How we work

• 
  
• We ship at 80% and improve in the next iteration. Our users are on live infrastructure projects, so we learn from production.
• 
  
• We'd rather you read the paper than reach for the nearest library. Most of what we build doesn't have a library yet.
• 
  
• Feedback is direct and it's about the work. We move too fast to be precious about it.
• 
  
• High ownership. If you spot the problem, it's yours to fix.
•