Discovering algorithms no human would write.

Algorithmical Superintelligence builds agent swarms that invent, verify, and open-source new algorithms for physics, starting with quantum simulation and real-time quantum error correction.

Get in touchRead the thesis

01 — Thesis

Mathematics has Lean. Physics has reality.

AI is on the verge of transforming mathematics for a simple reason: proofs can be checked by machine. An agent can attempt a million ideas, and a verifier, not a human referee, tells it which ones are right. Verification turns raw compute into reliable discovery.

Physics has verifiers too: exact solutions in limiting cases, conservation laws and symmetries, hard benchmarks, and ultimately experiment. Wherever a physical problem can be stated formally and checked automatically, a swarm of agents can search a space of algorithms far larger than any research group could explore.

ASI usually stands for artificial superintelligence. We mean something more specific: algorithmical superintelligence, systems that design the algorithms the physical sciences run on.

The verifiers

  • 01

    Formal proofs

    Correctness guarantees written in Lean and checked by a proof kernel, not a reviewer.

  • 02

    Physical invariants

    Conservation laws, symmetries, and exactly solvable limits that any valid simulation must respect.

  • 03

    Hard benchmarks

    Logical error rates, latency budgets, and fidelity against ground truth. Numbers that settle arguments.

02 — Approach

A discovery loop that runs at the speed of compute.

We pick problems where success is measurable, then let agents iterate against the measurement.

  1. 01

    Formalize

    Turn a physics problem into a precise specification with an automatic checker.

  2. 02

    Search

    Swarms of agents propose, implement, and mutate candidate algorithms in parallel.

  3. 03

    Verify

    Every candidate is proven correct, tested against invariants, or benchmarked, automatically.

  4. 04

    Publish

    What survives becomes open-source code and a paper.

Agent swarms

Many coordinated agents exploring algorithm space in parallel and sharing what works.

Harnesses & verifiers

Problem-specific environments that score every idea, so the search never relies on taste.

Scalable infrastructure

Compute orchestration built to run millions of candidate evaluations per problem.

World models

Learned models of physical systems that let agents reason about an idea before simulating it.

03 — Research

Where we're starting.

Problems that matter, admit automatic checks, and where a better algorithm changes what is possible.

01

Quantum simulation

Simulating quantum many-body dynamics is among the most compute-hungry problems in science. We search for new simulation algorithms and verify them against exact solutions and conserved quantities.

02

Real-time error correction

A fault-tolerant quantum computer needs a decoder that reads error syndromes and corrects them within microseconds, indefinitely. We design decoders that are faster and more accurate, benchmarked by logical error rate under strict latency budgets.

03

Solvers, rebuilt by agents

Maximum-likelihood decoding is an integer optimization problem usually handed to commercial solvers such as Gurobi. Our agents rebuild these solvers from scratch, in the open, and specialize them to the problem.

04

Physical simulation, broadly

Many-body systems, materials, fluids: any domain where a simulation can be checked against physics is a candidate for agentic discovery.

04 — Open science

We publish. We open-source.

An algorithm is only as valuable as the trust placed in it. Every result we stand behind ships with code, a paper, and the verifier that checks it, so anyone can reproduce it.

05 — Team

Led by a physicist who ships quantum software.

Portrait of Rafael Haenel

Rafael Haenel

Founder & CEO

Physicist and quantum error correction engineer. Rafael builds software for fault-tolerant quantum computers, including Tsim, a fast universal simulator for quantum error correction, and brings experience developing QEC software at QuEra Computing.

His research spans quantum error correction, topological phases, superconductivity, and quantum simulation on noisy quantum hardware.

citations
600+
h-index
15

Published in Physical Review Letters · Nature Communications · Science Advances

Selected work

Google Scholar

We're assembling a small founding team of physicists, ML researchers, and engineers. If you want to build superintelligence for physics, get in touch.

06 — Contact

Let's talk.

We're building the team and the compute to make algorithmic discovery routine.

rafael@algorithmical.si