About
An independent lab for mathematics and machine intelligence.
Lobsenz Research exists on the bet that the next generation of AI systems will be built by people who can prove things: about numbers, about models, and about the software that runs them. We do research in public, verify what we claim, and put the methods to work for organizations that need AI to be dependable.
Proof over persuasion
Theorems are machine-checked where we can manage it, experiments are preregistered, and evaluation protocols are fixed before results come in.
Open by default
Papers go to arXiv and code goes to GitHub. A result that others can't inspect and reproduce doesn't count yet.
Research that ships
The techniques we study, such as efficient inference, reliable agents, and careful evaluation, are the same ones we deploy for clients.
Leadership
Founder.
Lobsenz Research is led by Abraham Lobsenz, a mathematician and machine-learning researcher at Dartmouth College and a Jack Byrne Scholar in Mathematics.
Abraham's number-theory work, three papers in 2026, resolves the generalized Visibility Density Conjecture for nonzero polynomials, with the main theorem formally verified in Lean. In machine learning, Abraham works with Dartmouth's cognitive science faculty on how language models reason, running 25+ open-weight models through behavioral tests and causal interventions.
On the engineering side, Abraham built data infrastructure at Interactive Brokers, moving production pipelines from Perl to Python and cutting query latency from minutes to seconds. The work also includes software in daily production use in legal operations, and open-source scientific software used by a USC research lab for over two years. In quantitative finance, Abraham's team placed in the top 1% of 18,803 teams worldwide in IMC Prosperity 4.
- 2026 — nowAI/ML research, Dartmouth Cognitive ScienceReasoning and working memory across 25+ open-weight LLMs; PyTorch, CUDA, RunPod, Discovery cluster.
- 2025 — nowMathematics research, DartmouthLattice-point visibility: three papers on arXiv, one formally verified in Lean 4.
- 2026 — nowProduction systems for legal operationsmacOS case-filing app on the Clio API; mail-infrastructure migration.
- 2025Software engineering, Interactive BrokersPerl → Python pipeline migration; 50+ production scripts; REST API for real-time analytics.
- 2023 — 2024Research software, USC Mann School of PharmacyBuilt CAP, open-source confluence analysis still in use by the lab.
Work with us
Let's talk.
Research collaborations, applied-AI engagements, or a hard problem you'd like a second pair of eyes on.