# Datacurve

> Datacurve builds expert-created training data, reinforcement-learning environments and evaluations for foundation model labs and AI developer-tool companies. Its Shipd platform pays contributors for difficult technical work; its public DeepSWE benchmark tests coding agents on original, long-horizon engineering tasks.

- **Founded:** 2024
- **Headquarters:** San Francisco, United States
- **Founders:** Serena Ge (Co-founder and CEO), Charley Lee (Co-founder)
- **Products:** Shipd, Expert coding datasets, Reinforcement-learning environments, Agent trajectories and supervised fine-tuning, Off-the-shelf datasets
- **Notable:** Selected for Y Combinator Winter 2024., Announced US$17.7 million in seed and Series A funding by October 2025., Distributed US$1 million in contributor bounties by October 2025, according to the University of Waterloo.

## Products & services

- **Shipd** — Contributor platform offering paid technical quests in software engineering, machine learning and data science.
- **Expert coding datasets** — Custom training data for debugging, refactoring, performance optimization, code editing and design-to-code tasks.
- **Reinforcement-learning environments** — Realistic technical environments with long horizons, tools and domain-specific evaluation.
- **Agent trajectories and supervised fine-tuning** — Expert demonstrations and traces recording tool calls, checks, pivots and recoveries.
- **Off-the-shelf datasets** — Prebuilt, reviewed datasets structured for use in training pipelines.
- **Benchmarks and evaluations** — Custom evaluations of task-specific model capabilities.
- **DeepSWE** — Public benchmark of 113 original long-horizon engineering tasks across 91 repositories and five languages.

## Achievements

- Selected for Y Combinator Winter 2024.
- Announced US$17.7 million in seed and Series A funding by October 2025.
- Distributed US$1 million in contributor bounties by October 2025, according to the University of Waterloo.
- Founders recognized on Forbes 30 Under 30, as reported by Waterloo.
- Released DeepSWE and published benchmark artifacts and evaluation trajectories in 2026.

## Latest updates

- **2026-05** — Published the DeepSWE launch article on May 26; the corporate research index lists May 18.
- **2026-06** — DeepSWE v1.1 article dated June 14 introduced isolated patch grading, structured test reports and environment fixes.
- **2026-07** — DeepSWE paper posted to arXiv on July 8.
- **2026-09** — Official DeepSWE v1.1 page displays leaderboard updated September 22; GitHub organization also maintains Pier, its evaluation infrastructure fork.

## Links

- Website: https://datacurve.ai/
- LinkedIn: https://www.linkedin.com/company/datacurveai
- Twitter/X: https://x.com/datacurve
- GitHub: https://github.com/datacurve-ai

---

Profile page: https://yespress.io/datacurve
Published by YesPress — https://yespress.io
Last updated: 2026-09-30