In this talk we will present how LLVM-IR-Dataset-Utils can be leveraged to build LLVM IR-based datasets, how these datasets can be leveraged for the development of data-intensive machine-learned heuristics, and how readily extensible our utilities are to new data-driven use-cases inside of the LLVM ecosystem to inform the design of future optimization heuristics. We anticipate the datasets constructed with this tooling, and expansions thereof, to have the potential to drive future heuristics validation, correctness testing, machine learning for compilers, and compile-time performance tracking.