YC-backed Datacurve raises $15M to help AI models think, code, and reason better — TFN

Datacurve, a startup born from Y Combinator, has closed a $15 million Collection A spherical led by Chemistry. Present and new backers additionally joined, together with engineers from DeepMind, Vercel, Anthropic, and OpenAI.

This follows a $2.7 million seed spherical earlier, bringing Datacurve’s complete funding to about $17.7 million.

The brand new funding will assist the corporate to develop the workforce, enhance the platform, and scale the developer contributor system to gather extra complicated and precious datasets.

What Datacurve Does?

Co-founded by Serena Ge and Charley Lee, Datacurve focuses on constructing high-quality datasets for software program and AI growth, particularly for code.

The corporate focuses on post-training and analysis knowledge, providing options like supervised fine-tuning (SFT), reinforcement studying environments, and reinforcement studying with human suggestions (RLHF).

The corporate’s platform produces coding challenges, debugging duties, agent workflow traces, and personal repository benchmarks to assist AI fashions carry out higher in real-world eventualities.

Its bounty-based system — Shipd, engages high engineers to contribute knowledge by structured, research-grade challenges, guaranteeing accuracy and variety at scale. To this point, it has distributed over $1 million in bounties, reports Techcrunch.  

“We deal with this as a client product, not a knowledge labelling operation,” Ge stated to TechCrunch. “We spend loads of time desirous about: How can we optimise it in order that the individuals we would like have an interest and get onto our platform?”

By combining human experience with scalable infrastructure, Datacurve helps main AI labs and enterprises improve their mannequin reasoning, problem-solving, and coding capabilities, positioning itself as a significant knowledge associate within the quickly evolving AI growth ecosystem.

The corporate’s purpose is to make the information assortment course of participating and rewarding sufficient in order that proficient software program engineers wish to be a part of.

As AI fashions develop into extra superior, the necessity for higher knowledge grows. Easy datasets are now not sufficient. Datacurve goals to fill gaps for post-training knowledge, datasets required after a mannequin’s preliminary coaching to enhance efficiency in real-world duties.

Whereas Datacurve presently focuses on knowledge for software program and AI (code), its founders imagine the identical mannequin might develop into different domains, corresponding to finance, advertising and marketing, or healthcare, sooner or later, reviews TechCrunch