Building deterministic systems on top of probabilistic models

I spent the last decade in public markets deploying ~$1B in capital. One thing that always struck me: a surprising amount of core market infrastructure still runs on manual work.

Critical workflows start from messy source information that has to be interpreted, cleaned, and reconciled before it becomes usable. Entire teams exist just to bridge that gap.

A small group of engineers and domain practitioners started building Auxage to see if this could be solved with a different architecture.

So far we’re generating human-analyst–grade outputs across the S&P 500 with ~99% accuracy, already outperforming tools like Claude and several large incumbents on this workflow.

Interestingly, the hard part hasn’t been model capability — it’s data architecture and system design.

Curious if others here have worked on systems that turn messy real-world information into reliable, high-accuracy infrastructure. If helpful, happy to share more about what we're building at Auxage.