Inside the Model Factory — Eiso Kant, Poolside AI
In an interview on the Latent Space podcast, Eiso Kant, co-founder and CEO of Poolside AI, details the company's strategy for developing AI models tailored to software engineering. Poolside trains its own foundation models from scratch, prioritizing code-specific data over general text. Kant explains that the company invests heavily in data curation, filtering, and synthetic data generation to improve model performance on coding tasks. He discusses the technical challenges of training at scale, including infrastructure decisions and the need for efficient data pipelines. Kant also touches on the importance of evaluating models on real-world software engineering benchmarks rather than generic NLP tasks. The conversation highlights Poolside's belief that specialized models trained on high-quality code data can outperform general-purpose models on developer workflows.
Poolside's focus on code-specific training data and models may influence how developers evaluate AI tools for coding.