UCLA undergraduate in Data Theory, Cognitive Science and Political Science, class of 2028. I work on reliable AI for human and multi-agent decision systems, and I also build interactive simulations.
- proofnet-ir — verified correctness checking for proof nets in multiplicative linear logic, in Lean 4. v0.10.0 decides correctness with Guerrini's sequential procedure, with proved termination and a quadratic cost bound; 1,191 audited declarations use only Lean's three standard axioms.
- proof-graphs — order redundancy of Lean tactic proofs: preregistered experiments continuing ProofNet-IR.
- Human–AI grading (UCLA Political Science, in progress) — generative models as both student and grader, measured against a human baseline.
- MCM2026ProblemB — MCM 2026 Problem B, Finalist and MAA Award: MILP and Monte Carlo for lunar logistics with space elevators.
- Simplification-of-Arifovic-Strings-Simulation — an interactive simulation of a minimum-effort coordination game with death–birth dynamics, built for PS 189.
Each one runs in the browser with no build step. All of them, with source and tests, live in one repository, fushanbobfan.github.io, and the full gallery is at https://fushanbobfan.github.io/.
| ripplefield | a 2D ripple tank: interference, diffraction, reflection, refraction |
| gravity-garden | N-body gravity and orbital mechanics |
| murmuration | flocking: tune the steering rules, watch a flock emerge |
| epicyclon | draw a shape, watch rotating circles retrace it via the DFT |
| pathlight | grid pathfinding algorithms, step by step |
| chromalens | colour-vision-deficiency simulation for images |
- veridone — an evidence-first acceptance desk for AI-assisted software work.
- ai-eval-micro-lab — small, auditable experiments for evaluating AI model outputs.
- quant-research-micro-lab and useful-automation-lab — transparent tools for quantitative research and for repetitive local tasks.
- parquet-drift — catch schema drift across Parquet shards before readers do.