Q-Force
CACC 1ST PLACEHybrid classical + quantum-inspired graph ML system for cross-asset financial markets — MAPE cut from 7.8% to 4.9%, crash recall lifted from 68% to 82%. 1st place, CACC Coder's Arena 2025.
Origin
Q-Force started as my Semester 5 Graph Machine Learning elective miniproject and outgrew the assignment. The question: can a hybrid of classical forecasting, graph learning, and quantum-inspired computation predict cross-asset market behavior better than any one approach alone?
The system
Markets aren’t independent time series — assets move each other. Q-Force models them as a cross-asset graph and runs Graph Attention Networks over it, so each asset’s prediction attends to the assets that actually influence it.
That graph layer is combined with two other engines. Classical forecasting — Kalman filters, particle filters, and GARCH — handles the well-understood linear and volatility dynamics. Variational Quantum Circuits, run as quantum-inspired computation, model the non-linear dependencies the classical stack misses. On top sits a real-time crash-risk score built on Extreme Value Theory, which is concerned precisely with the tail events where standard models fail.
Who built what
This was a two-person project and the split was clean: I owned the graph ML and the quantum-inspired modeling — the cross-asset graph, the GAT architecture, and the VQC design. My teammate built the real-time streaming layer and the FastAPI + AsyncIO dashboard that serves predictions every 2 seconds, plus the integration between the two halves. Dr. Amruta Aphale at MIT-WPU mentored the project.
Results
We benchmarked classical, quantum-inspired, and hybrid variants against each other, and the hybrid won on every axis: MAPE fell from 7.8% (classical) through 6.2% (quantum-inspired) to 4.9% (hybrid); crash recall rose from 68% through 74% to 82%; simulated portfolio gain improved from +4.2% to +9.0%.
Q-Force took 1st place in the Data Design & Web Development track at CACC Coder’s Arena 2025.