Crystal Forge

Inspiration

Crystal Forge came about when one of our team members, a physics major, identified a real problem in quantum research: lattice simulations require an enormous number of computations, and most of that cost is wasted. He realized that with machine learning and smart software design, we could build a system that routes problems to the right solver and focuses measurements only where they matter most, cutting cost without cutting corners.

What it does

Crystal Forge decides whether a quantum lattice simulation needs classical or quantum computing, then smartly allocates limited quantum measurements toward the observables that matter most for superconductivity research, saving time and resources without sacrificing scientific accuracy.

How we built it

We built Crystal Forge with a React/Vite frontend and a FastAPI backend. The backend handles physics simulations, ML-based routing (CorrMap), and our measurement planning engine (QProbe). Everything is organized into clean layers: problem specification, solver logic, and physics-aware decision making.

Challenges we ran into

Generic black-box measurement planning just didn't cut it for the hardest quantum problems. We had to completely pivot and rebuild our measurement planner around superconductivity-specific physics channels, which turned a failing system into one that actually works where it counts.

Accomplishments that we're proud of

Our routing model (CorrMap) hits 85% accuracy at classifying when a problem needs quantum treatment. Our superconductivity-aware measurement workflow more than tripled hard-sector coverage compared to naive approaches. We're proud that physics-structured thinking beat generic ML.

What we learned

Generic solutions lose to domain-specific ones. When our black-box planner struggled, leaning into the physics of superconductivity unlocked real progress. The lesson: know your problem deeply before reaching for a general-purpose tool.

What's next for Crystal Forge

Tighter frontend integration for the full workflow, better UI explanations of quantum measurement results, and eventually replacing our Pauli-bundle approach with reduced density matrix methods, aligning us with cutting-edge quantum materials research.