Hermes is my self-hosted research backbone running on Ubuntu Server, built to make trading and experimentation reliable, fast, and reproducible.

At its core, Hermes centralizes market datasets and application state as a mixed storage layer: SQL for transactional/relational workloads (e.g., journaling, metadata, observability) and Parquet for large research datasets and efficient offline analytics. The storage layer is exposed for remote access so tools can read/write data consistently across machines.

Hermes also acts as the runtime platform for my live services -- most notably Pallas and Arxiv Sanity Preserver -- along with supporting modules such as data collectors and NLP components. Everything is containerized with Docker to isolate dependencies, simplify deployments, and keep upgrades controlled.

Reliability is treated as a first-order requirement. The system includes operational controls for access monitoring and process/resource supervision, plus service-to-service security (e.g., mTLS) to harden internal traffic and reduce blast radius. The result is a stable home lab that behaves like a small production environment for research and trading.