Fireworks Forge Brings Specialized Intelligence Into One San Francisco Build Room

Fireworks Forge is bringing the open-model conversation into Pier 27 in San Francisco on November 3. The in-person conference is sold out, but Fireworks is opening the keynote and selected sessions through a livestream beginning at 9:00 a.m. PT. That gives the wider AI community a window into a room built around a practical question: what changes when a company stops treating intelligence as a general-purpose layer and begins shaping it around its own data, domain, cost structure, and customers?

The conference is organized by Fireworks AI, with CEO and co-founder Lin Qiao opening the program before a fireside conversation with NVIDIA president and CEO Jensen Huang. Microsoft executive vice president Jay Parikh, Fireworks CTO and co-founder Dima Dzhulgakov, Genspark CTO and co-founder Kay Zhu, Replit president and head of AI Michele Catasta, Reflection AI CEO Misha Laskin, Mercor CEO and co-founder Brendan Foody, Meta senior engineering director Junbiao Tang, and Hugging Face co-founder and chief science officer Thomas Wolf are among the published speakers.

The Conference Moves Beyond the Model Menu

Forge is organized around specialized intelligence built on open models. The agenda moves through model routing, domain-specific training, inference for agents, evaluation pipelines, sovereign data, enterprise governance, and the economics of AI spend. Sessions from Fireworks, Genspark, UiPath, Doximity, Meta, PayPal, Replit, Notion, Zillow, Mercado Libre, Salesforce, and others put the technical work beside the operating environments expected to use it.

That mix matters because the next AI decision is increasingly less about finding one universally best model. Teams are deciding where intelligence should run, what data may reach it, how performance should be evaluated, which workloads deserve post-training, and who remains accountable when the system changes behavior in production. A conference can show the architecture. The companies on the agenda still have to make it hold across budgets, permissions, customers, and the engineers inheriting the system afterward.

What the Livestream Will Carry

The published agenda begins with Lin Qiao's keynote and the Jensen Huang fireside at 9:00 a.m. PT. The program continues across multiple tracks with sessions on enterprise automation, model routers, agent evaluation, inference efficiency, custom benchmarks, governance, and domain-specific systems. The livestream page specifically highlights Lin Qiao, Jensen Huang, and Jay Parikh, plus sessions involving Meta, Genspark, Replit, Reflection AI, and Mercor.

The virtual stream does not recreate every hallway conversation or workshop inside Pier 27. It does make the central argument available beyond the sold-out room: open models are becoming raw material for systems shaped around an institution's own work. The useful part will be seeing where the speakers draw the boundaries between model capability, infrastructure, application design, and the judgment that still belongs to the team deploying it.

Event Details

• Date: November 3, 2026
• Livestream: Begins at 9:00 a.m. PT
• In-person location: Pier 27, San Francisco
• Organizer: Fireworks AI
• Access: In-person attendance is sold out; virtual registration remains available through the official livestream page
• Register: Official Fireworks Forge livestream

Registration access, streaming availability, speakers, and agenda details can change. Fireworks controls the official event record and livestream access.