Market Decision Lab
A local experiment that runs Liquid AI d1-3B on real SEC filing excerpts to see whether a small model can classify warning signs and make bounded decisions.
A local research experiment that tests whether a small 3B model, Liquid AI d1-3B, can read genuine SEC filing excerpts and return a bounded decision. It runs fully on your machine, with no cloud AI and no accounts.
What it does
- Company Intelligence: revenue, net income, growth and net margin for Microsoft, Apple and NVIDIA, sourced from SEC XBRL records.
- Decision Inspector: classifies one filing excerpt on event type, reduced guidance (yes or no) and review urgency from 0 to 3, with probabilities and the raw model response.
- Portfolio Lab: set hypothetical weights across the three companies and see concentration measures.
How it works
- Financial numbers come from plain code and are never generated by the model.
- Every fact keeps its filing source, and every model answer comes from a real inference run.
- Stack: React and Vite frontend, FastAPI backend, llama.cpp serving the model.
Research and education only. Not investment advice.