Introduction

I build large-scale combinatorial optimizers that treat constraint spaces as thermodynamic systems — letting solutions flow to feasibility rather than search for it.

What this looks like in practice

• 80M clause university timetabling — 99.99999% satisfaction, 3GB RAM, single laptop core
• 2.6M clause hardware multiplier verification — 100% in 5.92 seconds
• Protein contact map prediction — 99.8% with zero domain-specific tuning
• Ramsey R(5,5,5) at N=52 — solved via thermodynamic fracture detection

These aren't curated benchmarks. Runtimes, RAM usage, and violations are all disclosed.

Projects

[NitroSAT](https://github.com/sethuiyer/NitroSAT) — Open-source O(M) MaxSAT approximator. Handles millions of clauses on consumer hardware. Free, Apache 2.0.

[BAHA](https://github.com/sethuiyer/baha) — Combinatorial optimizer using complex-plane branch enumeration (Lambert-W) to escape local minima. Outperforms SA/GA by 2–10× on phase-transition problems with zero tuning.

[Navokoj](https://navokoj.shunyabar.foo/) — Production API for large-scale MaxSAT, Max-QSAT, and XOR-SAT. Proprietary solver with global geometry awareness. Generous free tier.

Who this is for

If you work on scheduling, chip verification, logistics, graph theory, or biological constraints at scale , these tools are built for your problems.

Your sponsorship funds what comes next - GPU acceleration,
larger benchmark suites, and new problem domains. The core
will always be open. Sponsors make the research faster.