Fireworks opens the roof to AI engineers at AI Engineer 2026

Fireworks will host an open rooftop session on 13 October during AI Engineer 2026 in San Francisco. Billed as "Open Office Hours," the event has no slides and no pitches — just the Fireworks team and engineers who show up with laptops, questions, and production cases that are stuck. Anyone who wants to can leave at the end of the evening with a fine-tuned model deployed to a live endpoint on the platform.

The company describes itself as a platform for "specialized intelligence": not a single general model for everyone, but infrastructure that lets organizations train and serve models tailored to their own data, workflows, and use cases. Customers already running on it include Doximity, Revolut, and Shopify. The catalog spans hundreds of open models across text, image, embedding, and multimodal, served globally.

Behind the scenes, Fireworks was founded by the team that built PyTorch, the world's most widely used deep-learning framework. The investor list reads like a Silicon Valley who's who: AMD, Atreides, Benchmark Capital, Index Ventures, Lightspeed, NVIDIA, Sequoia Capital, and TCV. That backing lets the company run large-scale serving infrastructure and offer open models without customers having to manage GPUs themselves.

The invitation spells out exactly what to bring: a training run that won't converge, a prompt or agent loop stuck in optimization, questions on SFT, RFT, evals, LLM-as-a-judge, or deployment to live endpoints. "Hot takes" on open versus closed, routing, evals, or where the field is headed are also welcome. The question "Is training even worth it for my use case?" is explicitly listed as a legitimate starting point for discussion.

The organizers keep stressing the informal format: "a room full of AI engineers and tech leads building specialized intelligence, plus the Fireworks team, on a roof, with a good drink in hand." Someone who just wants a second pair of eyes gets one. Someone who wants to walk out with a tuned model deployed gets hands-on guidance with an open laptop. Registration is open through the link in the invitation.