Skip to content

Latest commit

 

History

History
221 lines (186 loc) · 17.5 KB

README.md

File metadata and controls

221 lines (186 loc) · 17.5 KB

Robin Langer — formally-verified mathematics, cryptography & trustworthy ML

Mathematician (PhD) and software engineer. I build systems whose claims a computer can check: machine-verified proofs in Lean 4, cryptography from first principles, interpretable and epistemically-honest ML, and autonomous research agents whose reasoning is adversarially verified. The through-line is rigour you can reproduce — no result below asks to be taken on trust.

Start with the two indices — the Lean / formal-verification index and the expository writing index — or scan the heroes and the category directory below.


The mathematical program — how the pieces connect

Much of the mathematics here is not a scatter of separate projects but one connected research program in algebraic combinatorics and representation theory, increasingly machine-checked in Lean 4:

  • Symmetric graphs ⟷ Hecke algebras. My thesis studies vertex- and arc-transitive graphs through the group acting on them. Every such graph is a coset graph Sab(G, H, D) (Sabidussi), whose connection set D is a union of H–H double cosets — precisely a basis of the Hecke algebra ℋ(G,H) = End_G(ℂ[G/H]), my doctoral area. Lorimer's theorem (arc-transitive ⟺ D a single double coset HaH) is the smallest non-trivial Hecke case. → lean-sabidussi (formalized, 0 sorry), symmetric-graphs (per-graph Hecke algebras + example catalogue), hecke-algebras.
  • q-series, symmetric functions & positivity. The same algebra one q-deformation over: symmetric functions and Macdonald polynomials (haskell, puzzles), cylindric partitions and Warnaar's positivity program (cylindric-partitions, warnaar-glue), with the umbral / finite-operator calculus as the Hopf-algebraic backbone (lean-hopf-algebras, umbral-calculus).
  • Formalization as the method. The graph theorems and the umbral/Hopf structures are written in Lean 4 and certified by #print axioms — claims a computer verifies, not claims you take on trust. This is the through-line to the cryptography and trustworthy-ML work: rigour you can check.

A parallel strand applies category theory to the compositional modelling of real systems — directed containers with a Lean-verified spine (macbeth-seed) and applied instances. Same habit of mind: find the structure, then make it formal.


Worth sixty seconds — the heroes

Ordered by what a high-assurance reader scans for first: verification, then cryptography, then trustworthy ML, then autonomous research, then range.

# Project The claim
1 · Formal verification lean-sabidussi My PhD theorem (Sabidussi representation + Lorimer, for symmetric graphs) machine-checked in Lean 4 — 0 sorry; the headline theorems reduce to Lean's foundational axioms, verified by #print axioms.
2 · Cryptography homomorphic-encryption Paillier homomorphic encryption + Schnorr zero-knowledge proofs, built from scratch with no dependencies.
3 · Trustworthy ML rsk-transformer · trust-boundaries A transformer learns the inverse RSK correspondence to 100% exact match (n=10) and 99.99% at n=15 — 0.00004% of the space seen, so algorithmic generalisation, not memorisation (a baseline MLP gets 3.07%). Paired with a formally-verified per-agent trust algebra.
4 · Autonomous research warnaar-loop-experiment A loop of amnesiac Claude agents — context cleared each round, notes the only memory — attacking an open positivity conjecture (Corteel–Dousse–Uncu / Warnaar). Across 27 agent instances it produced genuine partial results (a proof of the Q₁ case, a determinant identity for the Corteel–Welsh system, the Adjugate Monomial Theorem); the full conjecture remains open, and the repo says so.
5 · Range categorical-evolution · embeddings · smoke-flow A published GECCO 2026 result (categorical composition of GAs); a math-retrieval embedder that beats OpenAI's text-embedding-3-small (MRR 0.816 vs 0.461, reproducible benchmark); a real-time GPU Navier–Stokes smoke solver.

The two indices

And two program narratives — how the pieces fit together, not just what they are:


Category directory

Every public repository, grouped, strongest/most-relevant first. Claims are stated to match what the code actually shows.

1 · Formal verification & machine-checked proof — see the Lean index · program narrative

Repo What it is
lean-sabidussi PhD theorem (symmetric graphs ⟷ Hecke algebras), Lean 4, 0 sorry, axiom-checked.
lean-hopf-algebras Hopf algebras and the umbral calculus in Lean 4 / Mathlib (PARTIAL — honest split).
warnaar-glue Lean glue for Warnaar's A₂ cylindric-partition positivity; sorry-free.
trust-boundaries Per-agent trust-level algebra — zero axioms.
trajectory-tracing Epistemic-honesty formalism; sorry-free.
social-network Directed-container comonad laws (Haskell + Lean).
symmetric-graphs Per-graph Hecke algebras + example catalogue — companion to lean-sabidussi.
continued-fractions Legendre's palindrome / √n (STATED — honestly unbuilt).

2 · Cryptography, randomness & number theory

Repo What it is
homomorphic-encryption Paillier homomorphic encryption + Schnorr zero-knowledge proofs, from scratch.
watermark-rs Legendre-sequence spread-spectrum watermarking in the DFT domain (C++).
watermark-attack Diffusion-based attack pipeline vs invisible watermarks (Tree-Ring, Stable Signature, SynthID).
polynomial-primes Galois theory, irreducibility, Bunyakovsky — the algebra under public-key crypto.
category A Martin-Löf randomness test that pseudorandom generators provably fail.

3 · Trustworthy & interpretable machine learning

Repo What it is
rsk-transformer 100% exact inverse RSK (n=10), 99.99% at n=15; mechanistic-interpretability probes of how it computes.
octopus-streams Mechanistic interpretability of transformers via RSK (Pythia-70m).
bijection-transformer-interpretability SAE causal interventions; learning combinatorial bijections.
rsk-manifold-microscope The RSK transformer as a manifold microscope.
embeddings Math-retrieval embedder beating OpenAI text-embedding-3-small (MRR 0.816 vs 0.461).
clip-explorer CLIP zero-shot scoring with interactive UMAP.

4 · Autonomous agents & verified reasoning

Repo What it is
warnaar-loop-experiment A loop of amnesiac AI agents attacking an open positivity conjecture (Warnaar / Corteel–Dousse–Uncu); produced new partial results (Q₁ case, Adjugate Monomial Theorem). Full conjecture still open.
toy-agent-harness A minimal autonomous agent harness — essay + working code.
orchestration LLM topology-diversity experiments; a sign-flip result via contravariant Kan extensions.
mathematical-systems Why communication topology, not the fitness landscape, governs multi-agent diversity.

5 · Knowledge representation & retrieval

Repo What it is
description-logic-owl The logic under ontologies and automated reasoning.
torah-embed Talmud retrieval from a centuries-old hand-curated relevance apparatus (Sefaria data, CC-BY-NC).
embeddings Knowledge-graph-guided fine-tuning of an embedding model (see §3).

6 · Compositional systems modeling (directed containers) — see the container-theory program

Repo What it is
macbeth-seed The Lean-verified container spine (proven/stated split — see the Lean index).
category-of-containers The category of containers as a compositional foundation.
supply-chains · federation · org-chart · firm · stalin · voting · three-dates · dungeon One categorical idea applied across a supply chain, a federation, an org-chart, a firm, a famine economy, a vote, a dating scenario, a dungeon.
category-theory-you-already-know A programmer's on-ramp to category theory.

7 · Algebraic combinatorics & representation theory

Repo What it is
hecke-algebras My doctoral area — representation theory on the double-coset structure.
puzzles KTW puzzle triangles, Littlewood-Richardson coefficients, Hopf coproduct via transfer operators.
tableaux-mosaics Tableaux, mosaics, and q-analogues.

8 · q-series, symmetric functions & positivity

Repo What it is
cylindric-partitions Cylindric partitions and the Warnaar positivity program.
umbral-calculus Rota's finite operator (umbral) calculus.
haskell q-series algebra and symmetric functions — lazy power series, McIlroy streaming multiplication.
q-bool q-analogues of Boolean structures.

9 · Game AI & evolutionary methods

Repo What it is
nonaga A 14-weight genetic algorithm beats a 570K-parameter AlphaZero net 89% of the time.
connect_four AlphaZero wins head-to-head, but the GA plays better moves per a perfect solver.
blokus Blokus AI — GA heuristics, AlphaZero pipeline, web UI (collaboration).
checkers TD(λ) self-play, after Samuel's 1959 program.
categorical-evolution GECCO 2026 — GA operators composed as Kleisli morphisms; topology determines diversity.
evolve-evolution-strategy Meta-evolution of migration graphs on deceptive landscapes.
Topology-experiments Spectral analysis of migration topologies for island-model GAs (Haskell).

10 · GPU & creative coding — all live, see the demos index

Repo What it is
smoke-flow Real-time GPU smoke: laminar→turbulent via Navier–Stokes (live).
fire Real-time browser fire: buoyant Navier–Stokes + vorticity + combustion (live).
reaction-diffusion-screensaver Music-reactive Gray–Scott reaction–diffusion (live).
belt-trick The Dirac belt trick in 3D — why spinors need 720° (SU(2), Three.js) (live).
stomachion Archimedes' Stomachion — interactive exact-cover puzzle solver (live).
dancing · kaleidescope · hexstar · dmt · screen-saver More generative / geometric demos — all live (see the demos index).
sturmian-monster Sturmian words and continued-fraction structure (Haskell).

11 · Expository writing & essays — see the expository index

Repo What it is
kerry-questions Information-geometry and LLM-geometry essays.
pde-notes · langlands · ml-math-notes Applied-analysis, Langlands, and ML-mathematics notes.
medium-articles Essays on AI architecture and autonomous agent systems.

12 · Data & finance

Repo What it is
crypto Crypto-currency / commodities momentum trading (a bookmaker fair-price model).

13 · Books & long-form

Repo What it is
steel-book The Iron Thread — a 75,000-word narrative history of steel through civilizations.

14 · Published results & talks

Repo What it is
categorical-evolution GECCO 2026: "Composition Determines Diversity."
gecco-2026-talk Slide deck and per-slide narration for the GECCO 2026 talk.

15 · Infrastructure & tooling

Repo What it is
personal-assistant A CLI for mail, calendar, contacts and messaging — the pa/wa tools.

PhD in mathematics (symmetric graphs and Hecke algebras). Trained models are published as HuggingFace model cards: math-embed (math retrieval, MRR 0.816), torah-embed (Talmud retrieval), rsk-transformer (inverse RSK, 100% exact). Every headline result is verifiable — the Lean index tells you exactly what is machine-checked and what is merely stated, and the expository index is the prose behind it.