Ph.D. Candidate in Electrical and Computer Engineering at the University of Delaware
I build execution-contract systems across compilers, runtimes, and agentic AI for heterogeneous and distributed computing.
My current work focuses on making State, Dependency, and Effect explicit so scheduling, overlap, orchestration, and correctness become compiler-visible instead of ad hoc.

Location: Newark, DE, USA
Email: rafaelhg@udel.edu
GitHub: randreshg
LinkedIn: randres-herrera
X / Twitter: randres_hg
Portfolio: randreshg.github.io

I work at the boundary between compiler/runtime co-design and AI infrastructure. The recurring question behind most of my projects is simple:

How do we make execution semantics explicit enough that systems can schedule, validate, and optimize work automatically?

That question shows up in different forms across the repos I spend the most time on:

• Compilers and IRs for LLVM, MLIR, OpenMP, and typed workflow lowering.
• Runtime systems for event-driven execution, asynchronous overlap, distributed scheduling, and placement-aware execution.
• Agentic AI systems with contract-typed orchestration, graph execution, tool integration, and reproducible workflows.
• Research tooling for benchmarking, experiment tracking, and developer-facing infrastructure.

• Compilers: LLVM, MLIR, OpenMP, lowering pipelines, optimization passes, execution-model design.
• AI agents: typed orchestration, workflow graphs, execution contracts, tool runtimes, reliability.
• Runtime systems: task graphs, heterogeneous execution, distributed coordination, event-driven scheduling.
• Systems research: performance, correctness, reproducibility, validation, and developer productivity.

• apxm
  A-PXM: a parallel execution model for agentic AI with a graph IR, compiler pipeline, runtime, scheduler, and LLM/tool backends.

• agentmate
  A Rust framework for building AI agents and CLI applications with tools, streaming, memory, and APXM-backed execution.

• tully
  A local-first, domain-neutral experiment tracking and research context system for reproducible technical workflows.

• carts-benchmarks
  Benchmarks and harnesses for CARTS, focused on contract extraction and event-driven execution experiments.

• RayTracer_Javeriana
  Earlier hardware-acceleration work tied to my FPGA ray tracing thesis and systems-design foundations.

• AMD Research and Advanced Development: compiler/runtime co-design for heterogeneous execution.
• Meta: contract-aware agentic orchestration and privacy-aware infrastructure workflows.
• Argonne National Laboratory: OpenMP task-dependency graph discovery and exascale optimization work.
• Pacific Northwest National Laboratory: advanced-memory placement workflows for AI-science systems.
• Barcelona Supercomputing Center: OpenMP target-offload optimization for DPUs and heterogeneous pipelines.
• University of Delaware / CAPSL: compiler/runtime systems, teaching, and execution-model research.

• Execution contracts over State, Dependency, and Effect
• MLIR and LLVM pipelines for distributed and heterogeneous execution
• Reliable agent orchestration with typed IRs and compile-time validation
• Research infrastructure that is local-first, inspectable, and reproducible

If you work on compilers, runtimes, agent systems, or research tooling and want to collaborate, feel free to reach out.