Case Study: Selko.io automates identification of relevant sections in complex project documents with Valohai

Identify relevant text from complex documents

Selko.io helps multi-disciplinary project teams working from very large, multi-hundred-page project documents, where manually finding and allocating relevant sections to the right team members is a major burden. To solve this, Selko.io builds ML-based multilabel text classifiers using transfer learning on pre-trained models (e.g., Fast.ai, Hugging Face) and integrates with the Valohai platform — specifically calling the Valohai API — to orchestrate model training and inference on Selko’s AWS instances.

Valohai acts as the orchestration layer between Selko’s UI and their infrastructure: when users upload labeled examples (≈200 sentences per category) the Selko UI triggers training via the Valohai API, Valohai runs versioned training jobs and later serves inference, and the UI returns categorized documents to users. By using Valohai, Selko.io removed the need for dedicated DevOps, made experiments automatically versioned and shareable, sped up model iteration and deployment, and automated the previously manual document-review workflow.

Selko.io

Aditya Jitta

Senior Data Scientist