Sign in

SIH 26119 · Indigenous optimization

Submit on the web. Solve on your machine.

Upload LP, MILP or convex QP models. Route jobs to your workstation or GPU server. Get a verified result, not just a status line.

Open workspace How it works
Classes
LP · MILP · Convex QP
Compute
Your CPU or CUDA GPU
Connector
sovereign-compute

The problem

Optimization shouldn't mean a rented black box.

Refineries, power grids and logistics teams depend on mathematical optimization, but most either rent cloud solvers or rely on foreign licenses they can't inspect.

Sovereign answers SIH 26119 with a native C++ core, a web workspace, and a connector that runs on hardware you already own. The coordinator routes; the solve stays local.

Architecture

Model upload to verified result

Submit a job

Upload JSON or MPS. Pick algorithms, device, time limit, and target machine.

Coordinator routes

Backend queues the job and assigns it to a compatible online worker. No server GPU needed.

Machine solves

npm connector polls over HTTPS, runs Sovereign locally. No inbound ports required.

Result verified

Objective, variables, runtime, verification status, and actual GPU op count.

Website

Workspace for models, machines, results

Manage the full loop: create jobs, track queue progress, review solutions, control connected machines.

  • Firebase sign-in, no API keys
  • JSON and MPS upload or paste
  • Per-job algorithm and device choice
  • Live queued → running → done status
  • Approve 6-digit pairing codes
  • Disconnect machines anytime
Go to workspace →
sovereign workspace
JobMachineStatus
Blend planPlant srvDone
Unit commitGPU nodeRunning
Transport LPAnyQueued
$ npm install -g sovereign-compute@0.4.2

$ sovereign connect \
    --server "https://sovereign-we6b.onrender.com" \
    --duration 8

Pairing code: 482913
Connected · CUDA · 8h left

npm connector

One install. One command.

sovereign-compute bundles the Windows solver. No git clone or build step. Approve the code in your browser; pick 1–72 hour connection.

  1. Install globally with npm
  2. Run sovereign connect
  3. Approve code on the website
  4. Submit jobs to that machine

Solver core

Native C++ algorithms

Benchmark solvers compare results; they are not the production engine.

LP

Linear programming

Dual simplex with steepest-edge pricing on a sparse LU, and a Mehrotra interior point on a sparse LDLT factorization, with presolve.

MILP

Mixed-integer LP

Branch-and-bound / branch-and-cut with LP relaxations, cuts, heuristics and branching.

QP

Convex QP

Mehrotra interior point on a sparse quasi-definite KKT system, with Frank-Wolfe as a fallback and independent solution checks.

GPU

CUDA path

Interior point factors large dense systems on the GPU; the engine keeps sparse models on the CPU, where they run faster.

MIQP, NLP and MINLP are roadmap items per the problem statement.

Why Sovereign

See what the solver actually did

01

Solver independence

Core LP, MILP and QP logic in Sovereign code: inspectable, not a black-box wrapper.

02

Bring-your-own compute

Route to laptop, plant server or CUDA workstation. Skip cloud GPU rental per solve.

03

Honest GPU reporting

Results show if CUDA kernels ran and how many ops occurred, not just a "GPU" label.

04

Verification built in

Independent solution checks separate a returned answer from a validated one.

Measured

Benchmark evidence so far

0 Runs matching HiGHS, verified (Netlib, MIPLIB, industrial, scale)
0 Numerical errors in those 61 runs
0 Variables in a verified LP and QP solve
0 CUDA speed-up on a 1,500-row dense LP

Single runs on one laptop (RTX 2050), 30 s per job. The million-variable models are synthetic. The GPU helps only on large dense systems; sparse models run faster on the CPU.

Applications

Where optimization runs daily

RefiningBlending & scheduling
PowerUnit commitment
LogisticsRouting & planning
ManufacturingProduction models
ResearchBenchmark comparison
UtilitiesDispatch on owned infra

Get started

Three steps to first solve

  1. Sign in and open connect dialog
  2. Install sovereign-compute on your machine
  3. Submit a model: an example or your own file
Open workspace

Vercel frontend · Render coordinator · local solver