Building Reliable AI Agents Through Engineering + Governance
Author: Dr. Boyuan (Keven) Guan
Affiliation: FIU Library & GIS Center
First Public Release: November 8, 2025
Version: 1.1.0
DOI: 10.5281/zenodo.17561541
This repository serves as both a research workspace and a framework distribution hub for building reliable, trustworthy AI agents. We explore how AI agents can evolve through documented, reproducible, and production-grade workflows while maintaining operational safety and accountability.
New here?
- 🚀 Using AgentLoom Framework? → See agentloom-framework/
- 📚 Understanding the Research? → Start with docs/START_HERE.md
This repository contains:
A production-ready framework for building dual-role AI agents with fully-connected knowledge graphs. Build custom agentic AI systems in 2-4 hours.
📦 agentloom-framework/ - Complete framework package with:
- 9-phase protocol for agent development
- Pre-built builder components
- Specification-based generation system
- 95-98% success rate, 2-4 hour build time
Exploring the dual-helix architecture that combines engineering excellence with governance safety.
📚 research/ - Research materials including:
- Dual-helix architecture (Learning + Governance)
- Framework foundations and theoretical underpinnings
- Case studies and experimental validations
- Publications and academic positioning
Our research is built on the dual-helix concept - two interlocking systems that create reliable, trustworthy AI agents:
The Agentic-AI Engineering Framework enables knowledge accumulation:
Context Capture → Documentation → Indexing → RAG → Fine-Tuning
Result: Knowledge Growth & Continuous Learning
The co-agenticOS governance layer enforces operational boundaries:
Rules Engine → Coordination → Memory Boundaries → Verification → Adaptation
Result: Bounded Autonomy & Accountability
Verification points between helixes ensure learning occurs within accountable constraints, forming the foundation of Reliable Probabilistic Intelligence.
-
Navigate to AgentLoom:
cd agentloom-framework/agentLoom-v3/ -
Read the user manual:
- New projects: USER_MANUAL_NEW_PROJECT.md
- Existing projects: USER_MANUAL_EXISTING_PROJECT.md
-
Follow the 9-phase protocol to build your agent (2-4 hours)
-
Start with foundations:
- docs/START_HERE.md - Quick onboarding
- docs/framework-foundations.md - Theoretical underpinnings
-
Explore the dual-helix:
- docs/dual-helix-clarification.md - Engineering vs. Governance
- research/figures/dual-helix-diagram-spec.md - Visual specifications
-
Review case studies:
- docs/case-studies/ - Production validations
Agentic-AI-Research-Roadmap/
├── README.md # This file — Start here!
├── LICENSE # CC BY-NC 4.0
├── CITATION.cff # Zenodo citation metadata
│
├── agentloom-framework/ # 🎁 Framework Distribution
│ ├── agentLoom-v3/ # Latest stable release (v3.0)
│ │ ├── README.md # Framework overview
│ │ ├── AGENTLOOM_DESIGN.md # Design philosophy
│ │ ├── USER_MANUAL_NEW_PROJECT.md # Build new agents
│ │ ├── USER_MANUAL_EXISTING_PROJECT.md # Validate existing agents
│ │ ├── builder-assets/ # Pre-built components
│ │ ├── specs/ # Generation requirements
│ │ ├── examples/ # Reference patterns
│ │ └── phases/ # 9-phase instructions
│ │
│ └── [previous versions]/ # Version history
│
├── research/ # 🔬 Research Materials
│ ├── figures/ # Diagram specifications
│ │ ├── dual-helix-diagram-spec.md # Core architecture visual
│ │ └── memory-hierarchy-analogy-diagram.md
│ │
│ └── [additional research materials]
│
├── docs/ # 📚 Core Documentation
│ ├── START_HERE.md # ⭐ Quick onboarding
│ ├── framework-foundations.md # Theoretical underpinnings
│ ├── research-problems-and-positioning.md # Academic landscape
│ ├── dual-helix-clarification.md # Engineering + Governance
│ ├── agentic-collaboration-guide.md # Methodology (11K words)
│ │
│ ├── case-studies/ # 🔬 Production Validations
│ │ ├── README.md # Case study guide
│ │ └── dataverse-diva.md # Digital libraries validation
│ │
│ └── releases/ # 📦 Release Notes
│ └── v1.0.2.md # Latest release
│
├── CONTRIBUTING.md # How to contribute
├── CONTRIBUTORS.md # Recognition and attribution
├── DUAL_REPO_STRATEGY.md # GitHub (public) + GitLab (full)
│
├── .cursor/ # 🤖 AI Collaboration Rules
│ └── rules.md # Canonical references
│
├── meta/ # 🗂️ Project Metadata
│ ├── directory-index.yaml # Directory structure
│ └── search-manifest.json # Document collections
│
├── archives/ # 📦 Historical Materials
├── drafts/ # 📝 Pre-Publication Materials
├── experiments/ # 🧪 Experimental Code
└── temp/ # 🗃️ Working Documents
"LLM is the theorist, Agent is the practitioner."
To bring AI from conversation to production, we must systematize how agents:
- Accumulate knowledge from real tasks
- Document their decision-making processes
- Evolve through structured feedback loops
- Operate within accountable boundaries
- Persist context across interactions
This is achieved through the dual-helix architecture where engineering excellence (learning) and governance safety (boundaries) interlock at every stage.
Establish a repeatable, scalable, and evaluable Agentic-AI engineering framework that integrates LLM reasoning with traditional software engineering discipline and operational governance, enabling sustainable AI engineering where AI systems:
- Not only respond but operate
- Not only answer but record
- Not only reason but evolve
- Not only learn but stay accountable
We are validating this framework across multiple domains:
| Domain | Use Case | Framework Validation |
|---|---|---|
| Digital Libraries | Metadata repair, catalog enrichment (Dataverse) | RAG precision, knowledge graph evolution |
| GIS / Environmental Data | Buoy data annotation, anomaly detection (EnviStor) | Spatiotemporal reasoning, domain adaptation |
| Education | AI teaching assistants, research support | Adaptive learning, multi-role coordination |
| IT Operations | Log analysis, system self-healing | Feedback loops, autonomous decision-making |
| Industry Applications | Multi-agent coordination systems | Operational reliability, governance compliance |
We welcome contributions in multiple forms:
- Build agents for your domain
- Share feedback and improvements
- Submit case studies
- Validate framework in new domains
- Contribute theoretical insights
- Co-author publications
- Report issues and suggest features
- Improve documentation
- Share best practices
Quick Links:
- 📖 CONTRIBUTING.md - Contribution pathways
- 🤝 Case Studies Guide - Submit validations
- 🎯 Agentic Collaboration Guide - Methodology
- 🌟 CONTRIBUTORS.md - Recognition
This framework is part of a connected ecosystem:
| Repository | Role | Description |
|---|---|---|
| AgentLoom | 🎓 Research & Framework | This repository - research + AgentLoom framework |
| co-agenticOS | 🧠 Governance & Runtime | Operational standards, coordination protocols, safety boundaries |
[AgentLoom Framework] → Engineering helix: How agents learn and evolve
↓
[co-agenticOS] → Governance helix: How agents behave safely
↓
[Domain Applications] → Validates dual-helix in production
The dual-helix in action:
- AgentLoom (this repo) provides the learning/engineering strand
- co-agenticOS provides the governance/safety strand
- Together they create Reliable Probabilistic Intelligence
Stage 1 (2025 Q4): Framework workshop paper
Stage 2 (2026 Q1-Q3): Domain-specific case studies
Stage 3 (2026 Q4): Integrative system paper
Stage 4 (2027): Monograph and curriculum development
See docs/Research-Timeline-2025-2027.md for detailed milestones.
This Repository (GitHub - Public)
- AgentLoom framework (complete, ready to use)
- Framework methodology and conceptual materials
- High-level documentation and guides
- All
private/folders are excluded via.gitignore
Full Repository (GitLab - Private)
- Complete research materials including experimental data
- Detailed case studies with metrics and analysis
- All
private/folders with sensitive content
For details, see DUAL_REPO_STRATEGY.md.
| Type | Description |
|---|---|
| Framework | AgentLoom - Production-ready agent development framework |
| Software | Open-source tools and templates for agentic AI engineering |
| Publications | Peer-reviewed papers at PEARC, JCDL, ICSE, AI-Engineering venues |
| Education | Agentic-AI course modules and tutorials |
| Datasets | Curated agent-interaction logs for research use |
| Industry Pilots | Operational agents in production systems |
If you use AgentLoom or reference this research, please cite:
@software{guan2025agenticai,
author = {Guan, Boyuan (Keven)},
title = {AgentLoom Framework & Research: Building Trustworthy AI Agents Through Dual-Helix Architecture (Learning + Governance)},
year = {2025},
publisher = {Zenodo},
version = {1.1.0},
doi = {10.5281/zenodo.17561541},
url = {https://doi.org/10.5281/zenodo.17561541}
}For AgentLoom framework specifically:
@software{guan2025agentloom,
author = {Guan, Boyuan (Keven)},
title = {AgentLoom: A Framework for Building Dual-Role AI Agents},
year = {2025},
publisher = {GitHub},
version = {3.0},
url = {https://github.com/Keven1894/AgentLoom/tree/main/agentloom-framework}
}Dr. Boyuan (Keven) Guan
Lead Developer & Research Engineer
FIU Library & GIS Center
📧 bguan@fiu.edu
🌐 https://giscloud.fiu.edu/gis/about/faculty-and-staff/boyuan-guan/
© 2025 Dr. Boyuan (Keven) Guan, FIU Library & GIS Center
This work is licensed under a Creative Commons Attribution-NonCommercial 4.0 International License.
You are free to:
- Share — copy and redistribute the material
- Adapt — remix, transform, and build upon the material
Under the following terms:
- Attribution — You must give appropriate credit
- NonCommercial — You may not use the material for commercial purposes
We welcome collaborations with:
- Academic Institutions — Research validation and datasets
- Industry Partners — Real-world deployment and feedback
- Open Source Community — Tools, benchmarks, and standards
Current Partners:
- FIU Library & GIS Center (lead)
- FIU Computer Science Department
- FIU Division of IT
- Pelican, OSG, and other research infrastructure partners
This repository is a living research project. The AgentLoom framework is production-ready, while research components continue to evolve. Contributions, collaborations, and feedback are welcome.
Last Updated: December 1, 2025
AgentLoom Version: 3.0
Research Status: Active