Applied Compute builds Specific Intelligence for enterprises by training custom AI models on proprietary company data and deploying in-house agent workforces. Founded by former OpenAI researchers and backed by $80M from Benchmark, Sequoia, and Lux Capital, the company embeds engineers directly within client teams to build end-to-end training stacks, agent platforms, and continuous improvement tooling. Their deployments power agent workforces at Fortune 50 companies, with named customers including DoorDash, Cognition, and Mercor.
The technical stack centers on reinforcement learning infrastructure for training reasoning models, with engineering focused on post-training techniques, model validation workflows, and systems for rapid iteration. Applied Compute claims to validate and deploy models in days rather than months by building training and agent platforms entirely in-house. Their proprietary agents reportedly achieve state-of-the-art performance on customer evaluations, enabled by tooling that supports continuous improvement cycles while embedded with client engineering organizations.
The engineering environment prioritizes systems infrastructure for RL training at scale, with work spanning PyTorch, JAX, DeepSpeed, distributed systems, and GPU computing. Two-thirds of the team are former founders, and the roster includes researchers with prior OpenAI experience and International Math Olympiad winners. The company operates with a US headquarters and maintains an embedded working model where engineers build and operate training stacks and agent platforms directly alongside client teams.