# Applied Compute

> Applied Compute helps enterprises train, deploy, and improve custom AI models and agents. Its Applied Compute Agent Cloud (AC2) combines reinforcement-learning infrastructure, inference, observability, and applied research support, with customer work spanning DoorDash menu correction, Cognition code review, Mercor professional-task agents, and Harvey document analysis.

- **Founded:** 2025
- **Headquarters:** San Francisco, United States
- **Founders:** Yash Patil (Co-founder and CEO), Rhythm Garg (Co-founder), Linden Li (Co-founder)
- **Products:** Applied Compute Agent Cloud (AC2), Ari, Embedded applied research, Managed training and inference infrastructure, Applied Compute: Small
- **Notable:** DoorDash case study reports roughly 30% relative reduction in low-quality menus in a human-graded production A/B test., Cognition case study reports 10x faster bug detection than its frontier alternative, deployed in Windsurf Quick Review., April 2026 financing brought disclosed funding to $160 million at $1.3 billion post-money.

## Products & services

- **Applied Compute Agent Cloud (AC2)** — Unified open-weight model training, dedicated inference, evaluation, tracing, and continual-learning platform; announced in private beta on August 25, 2026.
- **Ari** — Applied research agent that inspects training runs, analyzes rollouts, and investigates failure modes and reward hacking.
- **Embedded applied research** — Researchers and engineers work alongside customer teams on datasets, graders, training, evaluation, and deployment.
- **Managed training and inference infrastructure** — Customer teams run experiments while Applied Compute manages infrastructure for model training and serving.
- **Applied Compute: Small** — Custom post-trained model reported in the Mercor APEX-Agents case study; benchmark results are specific to the reported evaluation.

## Achievements

- DoorDash case study reports roughly 30% relative reduction in low-quality menus in a human-graded production A/B test.
- Cognition case study reports 10x faster bug detection than its frontier alternative, deployed in Windsurf Quick Review.
- April 2026 financing brought disclosed funding to $160 million at $1.3 billion post-money.
- AC2 announced in private beta August 25, 2026.

## Latest updates

- **2026-10** — Published Ari reward-hacking monitor research on October 1, including an internal benchmark and human-review limitations.
- **2026-09** — Published open-weight deployment security guidance on September 24 and large-scale trace-analysis research on September 23.
- **2026-08** — Announced AC2 private beta on August 25; published Harvey Review Table training work on August 14.

## Links

- Website: https://www.appliedcompute.com/
- LinkedIn: https://www.linkedin.com/company/applied-compute
- Twitter/X: https://x.com/appliedcompute
- GitHub: https://github.com/Applied-Compute

---

Profile page: https://yespress.io/applied-compute
Published by YesPress — https://yespress.io
Last updated: 2026-10-01