ARC: Aerospace Regulatory Compliance

Team: Garren Huckaby

Project

Regulatory Compliance is a Huge Barrier for Small Aerospace and Defense contractors

  • Small to Midsize contractors are driving change in the industry
  • Complex regulatory compliance slows innovation and increases development costs by up to 150%
  • Agentic solutions provide an incredible new opportunity that can level the playing field
  • ARC accelerates the compliance cycle by processing external regulations locally, and ensuring internal compliance and traceability
  • Capturing 10% of small to midsize aerospace companies is a $10M market opportunity

System

System diagram

Methods

Customer-Focused Design

  • Working as a test engineering intern highlighted the need for a tool that was context driven, traceable, and secure
  • Talking with local businesses including CVM, Ducworks, Borsight, and EPS exposed a range of regulatory documents and use cases that weren’t immediately obvious to us
Screenshot of customer facing page

Requirements of the System

  • Adaptable – Using Next.js as a frontend, and Supabase as a backend/database allowed the software to be flexible, storing documents and requirements in customer designed workflows
  • Technical – Accurately interpreting technical, multi-modal documents required a system that could process the information in multiple ways (RAG, OCR, PDF coordinate mapping, vector embedded search)
  • Fast and Secure – The largest technical challenge was finding the balance between model compute, and storage capabilities (Current Demo uses OpenAI API, and Python FastAPI, local version uses Ollama and Gemma on SSD and runs off computer’s VRAM

Conclusion

  • • ARC clearly links internal documentation to a specific section of external regulations, verifying compliance. ARC also determines compliance gaps in your procedures
  • Getting the program to run locally at speed requires a delicate balance of model parameter count and computer VRAM capability
  • Many local companies have a need for document compliance tools, but do not always need the level of security to justify a specialized program
  • Hardware, not software is currently the limiting factor, and additional testing and refinement is needed to increase product value
  • As models improve, local, multi-modal AI systems will continue to develop