Mugen Giken Research support

Research support

Mathematical research with AI agents —
faster, and checked.

We help R&D teams and university research groups adopt AI-agent research workflows, take on contract research in mathematics and physics, and verify proofs and algorithms. Our founder holds a PhD in mathematical physics and has published, in full, the record of writing two papers with AI agents.

Book a free 30-minute call See the track record

Start with a fixed-price pilot

JPY 450,000 (excl. tax)2 weeks, one topic

Pick one of the following, and you get results in two weeks. Decide whether to continue after you see them.

  • Reproduce a paper or an in-house computation (numerics, simulation or derivation) and assess whether it fits your problem
  • Verify a proof or an algorithm (several independent audits; key steps formalized in Lean where feasible)
  • Prototype an AI research setup (one cycle on your topic, plus a first draft of the rules)

Deliverables: a report, the work and audit logs, and reproduction steps (including code and Lean files).

Sized to fit a research group's routine purchasing (procedures differ by institution). Paying from a research grant →

What we offer

Adopting AI-agent research workflows

Corporate R&D teams and university research groups

We show you how to split hypotheses, computation and proofs across dozens of AI agents, and how to catch their errors with independent audits, numerical cross-checks and formal verification — by running one full cycle on your own problem. We set up the environment and write the rules down for your team.

from JPY 300,000 Workshop (1 day)

from JPY 1,500,000 Hands-on support (3 months)

  • One real cycle on your own problem
  • Audit and cross-check rules written for your team
  • You do the research; we hand over the method and the setup

Contract research in mathematics and physics

R&D in manufacturing, energy, finance and similar fields

Deriving physical models, validating simulations, optimization, and proving that an algorithm is correct. A good first step: we reproduce a method from a paper and tell you whether it applies to your problem.

from JPY 500,000 Reproduce a paper and assess fit (2–4 weeks)

from JPY 3,000,000 Contract research

  • If we cannot solve it, you still get the record of what was tried and why it failed
  • Ownership and publication are set in the contract
  • Confidential data only goes to commercial AI services that do not train on it

Verification and formalization of proofs and algorithms

Teams working on cryptographic, control or financial algorithms; researchers before submission

Several independent audits look for errors and gaps in a proof. Key lemmas and computations can be formalized in Lean 4 / Mathlib so that a machine can check them.

from JPY 100,000 Verification of a proof or algorithm

from JPY 500,000 Formalization in Lean

  • You remain the author; our help is acknowledged
  • You get a work log that supports AI-use disclosure statements

AI research environment for your group

University research groups and corporate R&D teams

We build the environment for running research with AI agents inside your own machines and contracts: work logs, audit and cross-check rules, continuous checking with Lean, and data-handling settings — delivered as a package, with a walkthrough.

from JPY 1,000,000 Environment package (2–4 weeks)

  • You do the research; we hand over the environment and the rules
  • Built on commercial AI services that do not train on your data
  • Ongoing advice available on a monthly basis

Prices are indicative and exclude tax. We give a fixed quote after the free call. We work remotely from Tokyo, in English or Japanese.

Track record

The papers are under review at a journal and have not been accepted yet. Their checks so far are AI audits plus the author's own review; external peer review is still to come.

See the case studies (constants checked in Lean, a gap that slipped past every audit, a ledger of 452 failures) →

How we keep AI output honest

  • The writer is never the checker. Every claim is audited independently by a separate agent that starts with no context.
  • Every number is recomputed. Constants and estimates are checked with reproducible code, such as interval arithmetic.
  • The core is formalized. Key lemmas and computations are made machine-checkable in Lean.
  • Failures are logged. Abandoned approaches and the reasons go into a ledger, so the same mistake is not repeated.

Paying from a research grant

What paperwork can you provide?
Quotes, delivery notes and invoices, and supplier declarations your institution may require. We can also provide a sole-source justification describing what we do and why it is distinctive; the procurement decision stays with your institution.
Can grant money pay for this?
We expect to be engaged for a defined part of the work (computation, code, formal verification, literature review) through your institution's normal purchasing. We do not take over the core of a funded project. Please check eligibility with your grants office.
Can a larger job be split into smaller orders?
No. Splitting one job to stay under a purchasing threshold can break your institution's rules. For larger work we quote the whole job up front.
Can you deliver before the end of the fiscal year?
We commit to a start date and a delivery date at the outset. Tell us early if you need delivery within a fiscal year.

What we do not take on

We do not produce papers or proofs for someone else to publish under their name alone, and we do not work on theses, coursework or materials for hiring, promotion or grant reviews. Our involvement is always disclosed, as co-authorship or in the acknowledgements. Requests from students require written confirmation from their supervisor.

Our policy in full →

Start with a free 30-minute call

Tell us about your problem and where your research is losing time. If we are not a good fit, we will say so.

Email us