Overview
Engineering oversight for fiCurve, a financial modeling tool whose machine learning component prices new bond issues by learning from how previous issues were priced as market conditions change.
The founder had hired contractors — a frontend developer and a separate team handling model training and hosting across Google Cloud, Vertex AI, and GitLab — and needed someone technical enough to judge whether the practices in use would produce something usable. I reviewed architecture and process, set standards the subcontractors had to meet, and kept the build on a path toward a real product rather than a demo.
The structural problem this engagement solves is specific and underrated: a non-technical founder paying contractors has no way to distinguish reasonable progress from expensive motion. Everything a contractor reports sounds like progress. The fix is not more reporting — it is someone accountable to the founder who can read the actual work: the repositories, the training process, the deployment path, and whether the pieces will compose into a product.
The oversight ran from late 2022 into 2024, deliberately light-touch: enough review to keep standards holding, without becoming another layer the founder had to manage.