Back to Projects

The Heuristic Implementation of AI in Distributed Systems Engineering

This study demonstrates that AI is a high-precision power tool, not a replacement for expert engineers. Just as a nail gun makes a carpenter 10x faster, AI augments the operator by handling the manual labor of syntax and search, allowing the expert to focus on architecture and resilience. The core finding is that success in modern DevOps requires a "Human-in-the-Loop" (HITL) framework where the machine is used for the "What" (syntax) but relies on the human for the "Why" (resilience).

Core Technologies

ChatGPTGeminiCursorGitHub CopilotResearch

Architecture Components

  • Builds on a philosophy of using technology for community resilience and autonomy.
  • Connects technical work to a personal roadmap of creating a 'living laboratory' for off-grid tech and agriculture.
  • Aims to create a blueprint for others, including graduating foster youth.

The 'Logic Ceiling' & The Limits of Predictive Models

The primary challenge identified is the 'Logic Ceiling'—the threshold where an AI's predictive pattern-matching capabilities are superseded by empirical, real-world truth. AI models can hallucinate configurations for complex, novel problems.

  • During the Nomad Edge deployment, AI models incorrectly predicted 'Auto-Propagation' of routes for a manual Cross-Region VPC Peering setup in AWS.
  • AI prioritizes statistically likely patterns over documented truth, leading to confident but incorrect assertions for edge-case scenarios.
  • AI models often work with outdated information, described as "a fast runner on a map that is sometimes 12 months out of date."

The Manual Pivot: A Human-in-the-Loop (HITL) Framework

The resolution required a complete bypass of the AI's flawed logic. The proposed solution is a formal Human-in-the-Loop (HITL) framework where the engineer acts as the navigator.

  • Pivoted to manual research, sourcing 'Ground Truth' from primary sources like GitHub Issues and AWS Knowledge Center logs.
  • The engineer's role shifts from 'typist' to 'strategist,' using AI for speed on known tasks and their own expertise for navigating novel challenges.
  • The HITL model leverages AI for the 'What' (e.g., generating boilerplate Terraform syntax) while the human provides the 'Why' (e.g., designing a resilient network topology).

The Augmented Engineer & The Northern Blueprint

The thesis concludes that the most effective modern engineer is an 'Augmented Engineer' who treats AI as a pragmatic tool, much like a carpenter uses a CNC machine for precision cuts to focus on custom joinery. This philosophy extends to a replicable 'Digital Seed' for remote communities.

  • A 400% increase in deployment velocity was observed in the Nomad Edge project by using AI for toil and manual work.
  • The framework provides a blueprint for deploying localized 'Knowledge Hubs' in Northern and Indigenous communities that do not rely on fragile southern infrastructure.
  • Establishes a model for 'Blue-Collar Tech' that values pragmatic, resilient systems over clever, fragile ones.

Key Learnings & Decisions

The Toolbox: An Interface Comparison

  • **ChatGPT (The Multi-Tool):** Generalist brainstorming and translating complex concepts into "plain English."
  • **Gemini (The Connected Foreman):** Real-time research and verification against live documentation (Google, 2024).
  • **Cursor (The Specialized Workshop):** Deep, project-wide contextual understanding for hardcore building, though at a higher cost.
  • **GitHub Copilot (The Apprentice):** Automates "toil"—the repetitive, manual grunt work and boilerplate code (Beyer et al., 2016).

References

  • Amazon Web Services (AWS). (2025). VPC Peering Routing Documentation. AWS Knowledge Center.
  • Bender, E. M., et al. (2021). On the Dangers of Stochastic Parrots. ACM Conference on Fairness, Accountability, and Transparency.
  • Beyer, B., et al. (2016). Site Reliability Engineering. O'Reilly Media.
  • Department of Defense (DoD). (2021). Dictionary of Military and Associated Terms. Joint Publication 1-02.
  • Google. (2024). Gemini: A Family of Highly Capable Multimodal Models. Technical Report.
  • IBM. (2023). The Quantified Value of Human-in-the-loop AI. IBM Institute for Business Value.
  • OpenAI. (2024). Model Capabilities and Limitations. Technical Documentation.
  • Pan, Y. (2016). Heading toward Artificial Intelligence 2.0. Engineering, 2(4).
  • Schlosser, K. (2026). The Nomad Edge Project: A Case Study in Distributed Resilience. Portfolio Research.
  • Zanzotto, F. M. (2020). Human-in-the-loop Artificial Intelligence. IEEE Intelligent Systems.

Implementation Milestones

A breakdown of the key tasks and milestones that brought this project to life.

Research Phase

Complete

Initial research and literature review

Analysis

Complete

Deep dive analysis and findings

Documentation

Complete

Final thesis write-up and conclusions