TRADING STACK
Proprietary tools, models and data pipelines
The brief
A small trading operation has an awkward tooling problem: the software it needs — reliable data ingestion, reproducible model runs, a way to inspect both — is either expensive, generic, or both.
The brief was to design and build the proprietary stack: custom data pipelines, automated algorithmic models, and the tools the team uses to watch them.
The approach
I treat the infrastructure as deliberately boring: idempotent ingestion jobs, clean schemas, runs that can be replayed and compared. The interesting risk belongs in the models, not in whether last night's data landed twice.
The stack is built in small increments alongside the research, so every tool earns its place by being used the same week it is written.
The outcome
The first generation of the tools and pipelines is in active development, running alongside the research rather than as a separate rebuild.
Working as a co-founder also means owning the parts a larger team would split: architecture, backend, data and whatever the week needs.
A co-founder's build: the research pipeline, algorithmic models and internal tools a small trading operation needs, assembled from an empty repository.
- Team
- Founding team
- Duration
- Apr 2025 — present
- Contribution
- Architecture, data, backend, tooling