Adam A. Holmes
I am a Research Engineer & Computational Physicist (Ph.D. Theoretical Physics, Cornell) focused on efficient AI systems and neurosymbolic architectures.
I build high-performance systems that navigate vast search spaces efficiently—whether sampling from a diffusion model, finding the ground state of a molecule, or computing the optimal move in a game. I deployed Transformer-based semantic search in production in 2018 (pre-BERT) and invented SHCI, a quantum chemistry method with 2,100+ citations.
AI Systems & Research
Current projects exploring efficient inference, neurosymbolic architectures, and multi-agent systems.
| Project | Description | Tech Stack |
|---|---|---|
| 🌀 Diffusion LLM | Diffusion language model from scratch, exploring efficient inference for discrete sequence generation on edge hardware (Jetson Orin Nano). | Python • PyTorch • Diffusion |
| ♟️ Neurosymbolic Chess | A neurosymbolic chess engine (Rust) that integrates MCTS with a state-dependent alpha-beta portfolio to solve the “cold start” inefficiency of pure RL. | Rust • MCTS • PyTorch |
| ⚛️ Arrow (SHCI) | High-Performance Quantum Chemistry Engine. The reference C++/MPI implementation of Semistochastic Heat-Bath CI. I currently lead maintenance and architectural extensions. | C++ • MPI • HPC |
| 🗺️ Multi-Agent Pathfinding | A navigation stack combining Conflict-Based Search (CBS) for global optimality with ORCA for local avoidance. | Rust • CBS • ORCA |
Foundational Research & Impact: High-Accuracy Quantum Chemistry
My Ph.D. research introduced a new family of quantum chemistry methods—Heat-Bath Configuration Interaction (HCI) (Holmes, et al., JCTC 2016) and its successor, Semistochastic HCI (SHCI) (Sharma, Holmes, et al., JCTC 2017)—that significantly advanced the state-of-the-art in high-accuracy electronic structure calculations. The core innovation was a paradigm shift in algorithm design, replacing inefficient “generate-and-test” approaches with a highly efficient strategy that deterministically identifies the most significant components of the quantum wavefunction.
These new methods enabled calculations on a scale previously considered intractable, allowing us to produce the first near-exact potential energy surfaces for fourteen low-lying electronic states of the carbon dimer (Holmes, et al., JCP 2017), a key benchmark system for multireference quantum chemistry, and the ground state binding curve of the chromium dimer (Li, Yao, Holmes, et al., Phys. Rev. Res. 2020), a grand-challenge problem that had remained outstanding for decades, effectively “closing a chapter in quantum chemistry.” Today, SHCI is recognized as a leading benchmark method and has been implemented in or interfaced with major quantum chemistry packages.
Contact
I’m always open to discussing new research ideas, projects, or opportunities. The best way to reach me is via email or on LinkedIn.