Become a sponsor to Sethu Iyer
I build combinatorial optimizers that treat constraint spaces as thermodynamic systems — solutions flow to feasibility instead of searching for it. You hand me a model and a deadline. I hand you a decision you can defend, with the assignment, violations, timing, and verification attached.
The Stack
| Project | What It Does | Status |
|---|---|---|
| NitroSAT | O(M) MaxSAT approximator. 116M clauses, 10.7 MiB RSS. 99.7% median satisfaction. | 🔓 Open |
| BAHA | Lambert-W branch enumeration. 2–10× over SA/GA on phase-transition problems. | 🔓 Open |
| Reify | Declarative D SDK. model.require(sum <= 100) → verified decision. 30+ domain encodings. |
🔓 Open |
| Navokoj | Production constraint API. 3.7ms median, 90.1% complete SAT. Beats OR-Tools on time-to-feasible. | 🔒 Sponsors get credits |
Who It's For
- Researchers — Reproduce benchmarks. Stress-test the physics. Full 8.79 GB CNF corpus on HuggingFace.
- Developers — Navokoj API credits, early engine access, direct line to the roadmap.
- Enterprises — Design partner program. Pilot on your hardest instance. contact@shunyabar.foo
Where the Money Goes
| Monthly | Milestone |
|---|---|
| $500 | Independent benchmark reproduction by an external lab. Kill log included. |
| $1,500 | Engineer #1. Bus factor → 2. |
| $3,000 | First 3 paid design partners (hospitals / logistics). Their instances. Their receipts. |
| $5,000 | Engineer #2. Bus factor → 3. |
Sponsorship Tiers
| Tier | Price | Perks |
|---|---|---|
| Signal | $5 once | Name in README. You heard the transmission. |
| Witness | $15 once | ↑ + 1.5× Navokoj CPU credits |
| Verifier | $50 once | ↑ + L4 credits + early engine access |
Current goal: $500/mo.
Funds external verification + Reify Python bindings + first design partner.
Compute Pricing (Sponsors)
| Hardware | Base | Per Minute |
|---|---|---|
| CPU | $0.01 | $0.02 |
| L4 GPU | $0.25 | $0.10 |
| H100 GPU | $1.50 | $1.00 |
Preview keys: contact@shunyabar.foo — scoped, time-limited, no plaintext.
Open vs. Closed
Open: Research solvers, docs, tutorials, CNF corpus, verifier scripts, reproducible science.
Closed: SUTRA production engine, hosted execution, billing, Navokoj infrastructure.
Open research builds credibility. Closed production funds the next decade. Both are deliberate.
Not For You If…
- You want a drop-in OR-Tools replacement with zero evaluation
- You only care about theory, not deployment
- You want a free, fully-featured production solver today
Perfect for you if you have a hard constraint problem that existing solvers can't crack fast enough, and you need certified results with audit trails.
Transparency
Every month, sponsors get:
- Shipped — what built, what broke
- Learned — failures, surprises, feedback
- Stuck — open problems where your input helps
- Next — updated roadmap
Public changelog: NitroSAT/CHANGELOG.md
One-person research lab. One salary. One laptop. 116 million clauses in 10 megabytes of RAM.
Your sponsorship doesn't just fund code — it funds the time to keep asking hard questions and keep sharing the results openly.
The core stays open. Sponsors make the research faster, the verification external, and the bus factor greater than one.
— Sethu
Would be able to hire interns to manage navokoj.
Featured work
-
sethuiyer/baha
Simulated annealing which uses thermodynamics of the landscape to escape local minima and works amazingly on discrete combinatorial problems
C++ 4 -
sethuiyer/NitroSAT
Unusually powerful linear-time MaxSAT approximator that consistently hits 99.5%+ satisfaction across diverse categories including Graph Coloring, Clique, and Ramsey instances.
C 2 -
sethuiyer/casimir-sat-solver
when boolean logic meets quantum mechanics
HTML 2 -
sethuiyer/multiplicative-pinn-framework
A Research Framework for Solving PDEs with Multiplicative Constraints
Python 1