AI systems engineer at Multiverse Computing. Turning research into measured, reproducible systems for real hardware.
- ProvenanceGuard — source-aware factuality verification for MCP-based LLM agents. Published on arXiv, poster presented at the Agentic AI Summit 2026 at UC Berkeley.
- NVIDIA/nvflow — contributed finance grounding verification (merged) to NVIDIA's end-to-end synthetic data generation and evaluation pipeline framework.
- LibreYOLO — Core ML export implementation (merged) for open-source object detection.
- Scaling Laws for Energy Efficiency of Local LLMs — paper artifact for CPU-only LLM/VLM energy-efficiency research. arXiv.
- TurboQuant CPU — CPU-focused quantization with modified
llama.cppbuilds and reproducible benchmark artifacts.
LLM inference · speculative decoding · agent reliability · computer-vision deployment · model compression · reproducible benchmarks