The context layer itself
How business meaning, lineage and trust signals get modelled, then exposed to both AI agents and the humans who have to trust them.
Atlan is building the context layer for enterprise AI. Six months, full-time, shipping production code to Mastercard, Zoom, Nasdaq, Unilever and hundreds of others. You will be joining while the architecture, the abstractions and the surface area are all live questions.
95% of AI pilots fail. Not because the models are weak, but because they have no idea what the data means.
Someone has to build the layer that fixes that. We are.
Definitions, semantics, ownership, trust signals, policy. The things that turn a warehouse full of tables into something an agent can act on without inventing an answer. This is early. The foundations are still being laid, which is exactly why an intern here writes code that matters instead of code that waits.
How business meaning, lineage and trust signals get modelled, then exposed to both AI agents and the humans who have to trust them.
Connecting to the platforms real companies run on, and handling the messy realities that come with them.
Reliability and evaluation for AI agents operating on enterprise data, where a confident wrong answer is worse than no answer.
Developer tooling and internal platforms that make every other engineer on the team faster than they were last month.
The space moves weekly. New problems appear faster than we can staff them, and you will be at the front of that.
Same code, same context, same customer conversations that full-time engineers get. From week one, not month three.
You like building things that solve real problems, and you care deeply about which problems you pick. There is a trail of things you made because nobody asked you to.
You have rewired how you read, write, debug and ship. Two dimensions matter here: how well you build with AI, and whether you have actually built AI systems. Agents, RAG pipelines, MCP servers, eval harnesses, things that run.
You move toward unclear problems instead of waiting for them to be defined. When the spec is missing, you write it. When the abstraction feels wrong, you say so out loud.
You would rather get a rough draft in front of a teammate today than polish something alone for a week.
All over Zoom. We are interested in how you actually think and build, not in rehearsed answers.
The projects you have worked on, past internships if any, and what you have been building most recently. Come ready to talk through what you built, why you picked that problem, and the calls you made along the way.
Your problem-solving and your experience working with AI in practice. We will go through your personal and academic projects, so have the code, demos or docs open and ready to share.
How you think, build and grow. What energizes you, how you work with peers, and what you do when an approach you trusted turns out to be wrong. We look for self-awareness and honesty about what you do not know yet.
Have your projects open. Code, demos or docs, whatever shows the real thing rather than a description of it.
Six months, full-time.
First week of September 2026.
Fully remote, with team offsites every few months so you actually meet the people you ship with.
Graduating from a B.Tech program in 2027, or graduated in 2026.
A No Objection Certificate from your college is required before you join.
A real shot at a full-time offer at Atlan after graduation.
This cohort is referral only. We are not running an open application for it, and there is no form at the end of this page.
Every candidate comes in through an Atlanian who has seen your work and is willing to put their name next to it. That is a higher bar than a resume screen, and a much better signal for both of us.
6 August to 18 August 2026. Referrals for this cohort close on 18 August, so start the conversation early rather than on the last day.