ValuationIQ
An AI analyst that argues the methodology and writes the thesis, but never does the arithmetic — every number is computed in deterministic Python and carries a clickable derivation.
- Python
- FastAPI
- SQLAlchemy
- pandas
- SciPy
An equity valuation platform built for a small analyst team. It takes financials from wherever they actually live — a ticker, a three-statement Excel model, a CSV, a 10-K PDF — normalises them through an alias dictionary of roughly 300 line items spanning US GAAP and IFRS, corrects units, flags earnings-quality anomalies, and then runs the full institutional stack: DCF, trading multiples, DDM, LBO with IRR solved numerically, M&A accretion/dilution, WACC, Monte Carlo, residual income, sum-of-the-parts, and a football-field summary.
The part worth looking at
The language model never performs arithmetic. It chooses methodology, calls
Python functions as tools, and writes the narrative — and that separation is an
enforced architectural rule, not a convention. Every derived figure is written
to a derivation_log with its formula and its inputs, so any number in the
output can be expanded into the calculation that produced it.
The second rule is that nothing fails silently. A line item the normaliser does not recognise is surfaced as a warning; it is never defaulted to zero and folded into a total. In a tool whose entire output is numbers, a quiet zero is worse than a loud gap.
Underneath: a provider-agnostic router puts Anthropic, OpenAI, Google, OpenRouter and local models behind one interface, with a shared tool-schema translator, so adding a provider does not touch the orchestrator.