Case Study: Two Hat Security achieves faster, scalable training for sexual-abuse material detection with Valohai

Real world example of deep learning sexual abuse material detection

Two Hat Security builds automated systems to detect sexual abuse material in hard‑to‑reach parts of the internet and faced operational challenges scaling and iterating complex deep learning models. The small team had been managing Amazon G2 instances and rotating intern accounts, which made provisioning, access control and reproducible training slow and error‑prone. To address this, Two Hat Security adopted Valohai to speed up model training and retraining and to handle resource and access management.

Valohai provided Docker‑based workflow orchestration (via Valohai YAML), elastic management of cloud and on‑prem hardware, isolated runs with automatic shutdowns, and an online interface for monitoring parallel experiments. The result was faster, more reproducible experimentation — Two Hat ran 30 hyperparameter sweeps on their first try, eliminated internal competition for GPU resources, reduced maintenance overhead, and moved from ad‑hoc notebooks to productionized training pipelines, all supported by ongoing Valohai customer support.

Two Hat Security

David Wang

Data Scientist