Curating the rush to apply foundation models to NP-hard problems

What it does This repository is a curated bibliography of recent papers exploring how foundation models—mostly large language and vision models—can tackle combinatorial optimization. It splits the literature into two tracks: research that repurposes existing general-purpose FMs as solvers, formulators, or algorithm generators, and research that builds new domain-specific models for operations research. Each entry is tagged by the classic problem it addresses, such as TSP, VRP, MILP, or job-shop scheduling, and classified by its contribution type.

The interesting bit The list acts as a real-time map of a field still figuring out its identity, with entries ranging from arXiv preprints to Nature and NeurIPS. The value is in the taxonomy: instead of a flat link dump, it distinguishes whether a paper uses an LLM to generate heuristics, automate mathematical modeling, benchmark reasoning, or explain solver behavior.

Key highlights

• Two-track coverage: off-the-shelf LLMs versus domain-trained foundation models.
• Problem tags spanning operations research staples: traveling salesman, vehicle routing, mixed-integer linear programming, scheduling, bin packing, and satisfiability.
• Contribution-type labels: Solution, Algorithm, Formulation, Benchmark, and Interpretability.
• Direct links to paper PDFs and, in many cases, accompanying code repositories.
• Venue tracking from preprint servers to top-tier ML and operations research journals.

Caveats

• This is purely a reading list; it offers no code, solvers, or reusable frameworks of its own.
• Discovery is limited to scrolling a single large markdown table; there is no search, filter, or structured query interface.

Verdict Researchers and practitioners watching the collision of generative AI with operations research will want this bookmarked. If you need a library to actually solve optimization instances, this is not it.