What we're building

IDP is TrafficStars' internal product. Engineering, Sales, Performance, Compliance and other teams run their work in it every day. It's internal, but the pace and product bar are those of a top Silicon Valley startup. We ship daily, users get new features within days, and most of the code is written by AI agents together with the people who need the feature.

We don't chase AI because everyone else does. We use it because it works. When a new model, agent framework or tool comes out, we try it the same week. If it's better, it's in production soon after. If not, we drop it and move on.

What's already live 40+ product areas, 260+ background jobs and 100+ scheduled automations, used daily across the company:

• An AI assistant that acts: voice and chat, with tools across the platform and errands it runs on a schedule.
• AI moderation of ad creatives: a second-opinion model, video and iframe checks, redirect tracing, fraud scoring.
• A BI chat over our analytics, plus automated spend reports and anomaly alerts.
• Agent workstations: anyone at the company can build a feature with Claude Code.
• Self-updating company knowledge: docs and tickets indexed for RAG, meetings transcribed and summarised.
• Every LLM call traced and costed.

Most of what we build doesn't have a ready-made playbook yet. We're writing it as we go.

What you'll actually do About half your time you'll be building, and the other half keeping the whole thing fast, safe and moving.

• Ship features end to end: schema, API, background jobs, UI, production. Often in days, not sprints.
• Build AI that does real work: agents and assistant tools, retrieval, streaming, evaluation, keeping token costs sane.
• Test new things, fast. A new model, an agent SDK, an MCP server: try it, measure it, decide, ship it or kill it.
• Plug AI into real data: Postgres, ClickHouse, internal APIs, Slack, email, LLM providers through OpenRouter.
• Run long and scheduled work on Trigger.dev: reports, checks, syncs, with retries that don't double-send.
• Be the last pair of eyes before production. You'll review PRs from several teams and their agents and catch permission gaps, data leaks, heavy queries and migration clashes.
• Make our coding agents smarter: skills, rules, CI checks. The better they get, the more the whole company can build.
• Ship it yourself: migrations, deploys to Kubernetes, and a rollback if something goes wrong.

You'll fit in if you

• Have 3+ years building real products, with full-stack work in production.
• Are genuinely curious. You read release notes for fun, you've already tried that new model, and you can tell hype from something useful.
• Communicate clearly. You turn a vague request from Sales or Compliance into a working first version, explain trade-offs to non-engineers, and give review comments people actually like getting.
• Own things. You don't wait for a perfect spec: you clarify, ship, listen and iterate.
• Have shipped LLM features to real users: tool calling, streaming, context limits, retries, caching, cost control.
• Think about AI safety in practice: prompt injection, keeping personal data out of prompts and logs, tools with only the permissions they need.
• Know how to tell if AI got better or worse: test sets, regression checks, not just vibes.
• Already work with AI coding agents daily (Claude Code, Codex, Cursor or similar). You know how to delegate to them and when not to trust them.
• Are strong in TypeScript, React and Next.js: you can design a feature from scratch.
• Are comfortable with Postgres: schemas, migrations, indexes, and why a query is slow. Supabase is a plus.
• Understand Kubernetes and Docker well enough to deploy, read a failing pod's logs and roll back. You don't need to be a k8s guru.
• Know the security basics: access control, injections, secret handling.
• Are fluent in Git, including messy merges between parallel branches.

Bonus points

• ClickHouse or other analytics databases.
• Observability: OpenTelemetry, Grafana, debugging through traces.
• You've built internal tools or developer platforms before.
• Ad-tech experience: advertisers, publishers, ad formats.

Why you might love it here

• Top-startup pace with real users from day one: a whole company relies on what you ship, and you never have to fight for traffic or a first customer.
• New AI tools land here first. You'll work with the latest models and agents, not read about them.
• Idea to production in days. You build it, ship it and see colleagues use it the same week.
• Your work multiplies. Every tool or agent improvement you make lets dozens of people build faster.
• A lot of autonomy and real influence on how AI is used across the whole company.