platoseedplatoseed
Frontier coding data for training and evaluating LLMs
We generate expert quality coding data at scale for fine-tuning LLMs
Datacurve positions itself as a data engine for frontier AI, focusing on custom data, evaluation, and research infrastructure to advance long-horizon reasoning, software engineering, and data science. It emphasizes data collection, rigorous annotations, and benchmarking to improve model capabilities.
Datacurve provides research and data collection infrastructure for frontier AI, including datasets and benchmarks (e.g., DeepSWE long-horizon coding benchmark) used to evaluate and iterate AI models. The platform centers on curated data, annotations, and evaluation to guide model learning and capability development across long-horizon reasoning tasks and software/data science workflows.
Who it’s for: AI research labs, academic research groups, and enterprises building or evaluating frontier AI models that require long-horizon reasoning, coding benchmarks, and rigorous data pipelines for model training and evaluation.
research infrastructure and benchmarking focus; mentions of products, research, and careers imply active development and recruitment
Started building software in high school - built a climbing training app with Team Canada athletes. Studied at Waterloo CS for a year then dropped out. Worked with the Cohere CTO on LLM reasoning and coding capabilities through synthetic data. Went to YC W24, pivoted 3 times until Datacurve. Now scaling high quality coding data production pipelines at Datacurve to enable next generation coding models
Hacking on things since middle school. Went to Waterloo CS, interned at Google, then dove into AI research on multi-modal RL and training browser-use agents. Went through YC W24, pivoted a few times, and landed on Datacurve – now providing the data infrastructure for frontier LLMs.
Providing code data by the best engineers, so you can build the most capable model
Datacurve provides expert-quality code data at scale generated by skilled software engineers. It serves AI dev-tool startups and foundation model labs with data for tasks like code completion, debugging, refactoring, and code generation to improve model capabilities.
From the original launch (Mar 2024) — may be outdated.
Formerly “Uncle GPT”, “Shipd” · why startups rename →
Frontier models for critical domains
AI Employees for the Enterprise
