We're going to transform education. Not through politics, but through engineering.
In 1984, Benjamin Bloom reported that students taught one-on-one with mastery learning performed roughly two standard deviations above students in conventional classrooms (paper). The challenge was making that kind of teaching available to every student.
Four decades later, we finally have a real shot at it.
We're building Bloom: an AI teacher students talk with, that explains and works through problems with them on a shared whiteboard. Bloom is live on Android. Early feedback from students and parents is encouraging. Much of the hard work, building a product that reliably helps children learn, is still ahead.
We raised $2.5M, led by Peak XV.
The problems
The model is not the product. Turning a language model into a teacher that actually produces learning is engineering, and there's no playbook for it yet. The questions we're working on right now:
How does Bloom decide what to say next: when to explain, when to ask, when to let a student struggle, when to step in?
What does the harness around the model need to look like for it to behave like a teacher, turn after turn, for a full session?
How do we know a student learned something, and not just that the session ended? If we can't measure learning, we're guessing.
What should the interface and the whiteboard be, when the user is a twelve-year-old and the teacher is talking?
How do you make voice, thinking, drawing, and replying feel like one conversation, not a phone tree?
These aren't solved. That's why we're hiring.
What you'll own
This isn't a role with a lane. You'll own problems like the ones above end to end: the agent, the evals, the product a student touches. As a founding engineer you build whatever Bloom needs to become a great teacher, and you shape how we build it.
Who this is for
2+ years building production systems of any kind. We care about the depth of your work more than what exactly it was.
You likely come from a top-20 engineering college, or a top Indian startup.
Fluent with LLMs: prompting, structured outputs, tool use, evals. You read traces and get annoyed at flaky ones.
AI coding agents (Claude Code, Cursor, or similar) are your daily driver, and you keep up with the frontier.
Bengaluru, in person.
You want to ship products that run in production and get used, not demos.
Bonus: you've built LLM or agent systems that were the core product of a company, not a feature bolted onto one.
Who it isn't for
You want a remote seat. This is an in-person job.
You want a clearly defined scope. Here the scope is whatever Bloom needs next.
Why this role exists
Every child who truly understands math becomes someone who can build, invent, and create what comes next. Nobody has been able to give every child a real teacher. We think we can.
If you're looking for work that's deeply meaningful, that makes the world better and lets you build something great for yourself, apply here.