70B Air-Gapped RAG on DGX Spark

Running 70B-parameter LLMs air-gapped on NVIDIA DGX Spark. No cloud. Just results. Prototypes include air-gapped 70B RAG on DGX Spark and edge RAG on Intel Lunar Lake Getac F120, focusing on distributed training and model parallelism.

GitHub Repo Need Help? DGX Spark nvidia-smi

About Shawn David Iuliucci

With over 20 years in software engineering, I've scaled large-scale systems to handle billions in value, from cloud to no-cloud environments. Now advancing large-scale AI frameworks, with a strong interest in contributing to NVIDIA’s Megatron Core and NeMo teams.

Proficient in Python, and actively mastering Rust and Kubernetes through hands-on projects. Early in specialized AI career but fully dedicated to scaling LLMs for real-world impact.

Building 4 open-source RAG tools across key domains, with 4 arXiv papers and 4 conference keynotes in the pipeline.

Key Skills & Expertise

AI Prototyping & Scaling (DGX Spark, LLMs)
Distributed Training & Model Parallelism
Air-Gapped & Edge AI Deployments
Python, Rust, Kubernetes
Large-Scale Data Processing (Millions of Files, Databases)
Web Development (Next.js, Cloudflare, Static/Dynamic Sites)
Content Generation & Automation (Runpod, Image Overlay)
Rugged Hardware Integration (Getac Models)
Open-Source Contributions & Frameworks

6-Month Domination Roadmap

Month 1: Defense

classified-rag: 70B on synthetic CUI, 0.9s latency

Month 2: Rail

rail-rag: 100k containers in 0.2s

Month 3: Finance

fin-rag: 1B trades in <5 min

Month 4: Healthcare

health-rag: 50 hospitals, federated, no PHI

Months 5-6: Integration & Leadership

All domains unified: 4 keynotes, DGX leaderboard