Product Engineer

In education, we have always traded off scale and quality. In a 30-student math class, half of kids are ready to accelerate while half are still working on their foundational skills, but teachers cannot be in 30 places at once, giving each student one-on-one support.

Until now. Imagine 30 teachers in every classroom—that's how Goblins feels. Students draw math on any device, and our AI figures out the "why" behind their confusion, giving instant feedback and building conceptual foundations.

We're backed by the Gates Foundation, and we're on track for 300,000 active students next year. We're hiring a product engineer to help us get there, and to make sure everything we've built holds up at that scale.

What you might work on:

• Scale our real-time teacher avatars to 300k students. Our Goblin avatars stream live to students in production today. Make them run smoothly at scale while cutting latency and cost, and think hard about reliability from every angle: Chromebook CPU, low-bandwidth classrooms, and the load spike when a whole district logs on at 9am. We really don't want to crash the school's network.

• Make model experimentation something an agent can run end to end. We're always testing new models against each other in real classrooms. Build the machinery to pin a model for a cohort, roll it out gradually, watch the metrics, and roll back fast, all clean enough that we can hand the whole experiment loop off to an agent and let it run.

• Keep every classroom in real-time sync at 300k concurrent students. Our real-time engine (think Linear) keeps every student's whiteboard and every teacher's dashboard instantly in sync. Make that hold up when thousands of classrooms are live at once, without melting our servers or the school's wifi.

• Give teachers instant, reliable insight across thousands of simultaneous classrooms. Teachers need to see who's stuck and what to assign next in real time, even at peak load. Build the systems that keep those live reads fast and trustworthy.

• Create shareable "aha moment" artifacts from real student sessions. Automatically find the best moments, strip out anything sensitive, and produce replays that teachers, parents, and districts want to watch and share, built to run cheaply over hundreds of thousands of sessions.

How we build: we're all in on agentic development. Our remote dev environments make it normal to have five or more agents running per person at any given moment, and we design our systems and workflows so work can be handed off to them cleanly. We'd rather you orchestrate agents than grind code out by hand.

Under the hood we're big on Effect-TS, Postgres and Bun + a real-time sync engine in the same spirit as Linear's.

The team is small and in person in Williamsburg: Sawyer (CEO, former Head of Design at X1, acquired by Robinhood, and a former math teacher) and Alp (CTO, ex-Stripe and Amazon). You'll be engineer #3.

We build software that figures out the "why" behind a kid's confusion. Show us you can do the same by decoding this :)

91 114 122 39 111 114 116 130 123 130 131 69 146 159 93 85 128 170 164 176 181 111 204 200 214 215 135 221 210 228 244 240 238 171 192 183 228 268 283 245 296 279 223 309 308 231 308 331 339 293 336 350 339 359 362 373 384 368 389 393 329 392 418 420 359 349 414 434 468 381 450 471 479 481 480 415 494 499 433 530 530 538 470 547 540 475 546 489 565 580 576 576 597 535 531 601 606 631 620 651 644 589 668 671 603 698 698 710 638 715 708 645 714 729 663 717 719 724 685 717 702