A valuation model built from free public data.
Institutional-grade scoring without institutional-grade data budgets.
Built entirely from free public data feeds.
Sector
Finance, wealth management
Tech
Python, pandas, statsmodels, free public data APIs (financial, macro, domain-specific)
The challenge
A wealth management firm wanted a valuation framework for an asset class they were already exposed to, built to the same rigour a serious research desk would use. They did not want to pay for institutional data feeds. The inputs they needed were available on free public APIs, but scattered, inconsistent, and in different units, distributions, and noise profiles.
Our approach
- —Data sourcing: stitched together a patchwork of free public APIs, each with different rate limits, coverage windows, and missing fields. Getting them into one consistent dataset was a meaningful share of the engineering.
- —Transformation: built the layer that converted each raw series into a bounded, z-scored, or percentile-mapped variable the model could actually combine without lying. Reconciled feeds that disagreed on timestamps, units, and coverage.
- —Optimisation: tested weight combinations, evaluated feature strength, and tuned the composite scoring formula to pull the strongest possible signal out of what the data allowed.
- —Handover as production-ready Python, designed to drop into the client's engineering stack.
The outcome
A working valuation model the client's engineering team integrated into their IT ecosystem. Six technical write-ups, one full scoring spec, working Python code. NDA in place, full IP with the client.
Where the real work was
The transformation layer. You cannot just average a handful of metrics that live on different scales. The final score is only as honest as the variable transformation layer underneath it. We wrote that layer to survive an audit.
Tags