Paul Richardson
Teacher | AI Engineer | Cloud Architect | Agentic Systems Builder
Maine, USA
https://www.oddlytrue.ai
paul@oddlytrue.ai
Professional Summary
I am a technologist, educator, cloud architect, and AI systems builder focused on making advanced technology practical, useful, and understandable.
My work sits at the intersection of AI, cloud architecture, agentic workflows, education, IoT, and human-centered software design. I build real systems, teach real students, and focus on tools that reduce friction instead of adding complexity.
Current work includes Halo Lumin, GuideCards / GuidePaths, Model Context Protocol integrations, AI-powered passive solar homes, local-first AI systems, and practical AI/cloud education.
My design philosophy is simple:
Build useful systems. Keep them clean. Make them understandable. Give people control.
Core Skills
- AI Systems: LLM integration, agentic workflows, prompt systems, RAG, local AI models, AI-assisted learning tools
- Model Context Protocol: MCP servers, scoped agent access, tool schemas, OAuth/DCR exploration, Codex and ChatGPT integrations
- Cloud Architecture: AWS, serverless systems, VPC design, S3, Lambda, DynamoDB, API Gateway, CloudFront, IAM, CI/CD
- Web Development: React, Next.js, Vite, Node.js, Flask, REST APIs, modern frontend architecture
- Platforms: Supabase, Vercel, Cloudflare, Docker, GitHub, Linux
- IoT & Edge Systems: Raspberry Pi, XBee networks, sensors, local automation, telemetry pipelines
- Teaching: Curriculum design, project-based learning, technical mentoring, AI/cloud instruction
- UX / Product Thinking: Simple interfaces, guided workflows, clean design, task-first software
Teaching & Education
Southern Maine Community College
Instructor – Architectural and Engineering Design
I teach technical and design-focused courses with a strong emphasis on real-world systems, emerging technology, and hands-on learning.
My instruction blends traditional design principles with modern tools including:
- AI systems
- Cloud architecture
- AWS services
- Raspberry Pi and edge computing
- Software development
- Technical problem solving
- Project-based learning
I focus on helping students understand not just how tools work, but how to think clearly while building with them.
AI, Cloud, and Developer Instruction
I also design and teach courses and workshops around practical modern technology, including:
AWS Developer & Cloud Architecture
- Secure AWS architecture
- Serverless application design
- Deployment pipelines
- Private networking and VPC design
- CloudFront, S3, Lambda, DynamoDB, API Gateway, and related services
- Real-world architecture decision making
AI Foundations
- AI fundamentals for beginners and intermediate learners
- Prompt engineering
- Local AI models
- Agentic workflows
- Retrieval-Augmented Generation
- AI-assisted software development
- Raspberry Pi and hardware-integrated AI projects
Codex, Agents, and AI-Native Development
I actively explore how tools like Codex, ChatGPT, MCP, and agentic workflows can help students, developers, and non-developers build useful systems faster.
My focus is not simply “AI writes code.”
My focus is:
Humans set direction. AI helps with execution. Good systems keep the human in control.
Featured Projects
Halo Lumin
Frictionless content-to-link sharing platform
Halo Lumin is a platform I built to make sharing content simple:
Content → Halo → Link → Share
Instead of sending attachments, users can turn text, Markdown, files, and structured content into clean shareable web pages called Halos.
Halo Lumin is designed around simplicity, low friction, and clear presentation.
Key Ideas
- Replace messy file-sharing workflows with clean links
- Make content easy to create, share, view, and revisit
- Keep the interface simple and calm
- Support both human users and AI-assisted workflows
- Make sharing feel lightweight rather than technical
Technical Stack
- Next.js
- React
- Supabase
- Vercel
- Serverless APIs
- Cloudflare
- MCP-ready agent access patterns
Halo Lumin Agent Access & MCP
Secure agent access for AI tools
I am building MCP and agent-access capabilities into Halo Lumin so tools like ChatGPT, Codex, and local developer agents can safely interact with Halo content.
The goal is not to expose everything to an AI agent.
The goal is to create a narrow, controlled, auditable tool surface.
Current Focus
- Model Context Protocol integration
- Scoped Agent Access keys
- Read and create permissions
- Safe tool schemas
- OAuth and Dynamic Client Registration exploration
- ChatGPT Developer Mode testing
- Codex CLI and developer workflow testing
- Fail-closed security posture
- Clear documentation for safe agent use
Design Principle
AI agents should have limited, explicit permissions.
They should not get broad access to databases, files, private data, admin tools, or anything they do not need.
GuideCards
A design model for AI-age software
GuideCards is my design framework for building software in the AI age.
Traditional apps often depend on menus, dashboards, forms, and buttons. GuideCards rethink this model around guided interaction.
The core idea:
One current card. One useful next step. AI-guided, but user-controlled.
GuideCards are not just buttons. They are focused units of guidance, action, explanation, reflection, or progress.
GuideCards Principles
- Keep the interface calm and focused
- Avoid overwhelming users with giant dashboards
- Let AI guide the workflow without taking control away
- Make each step understandable
- Keep users moving from intent to action
- Design for humans first, not just engineers
GuideCards can apply to education, onboarding, software workflows, support systems, training tools, and AI-assisted applications.
GuidePaths
An implementation of GuideCards for learning
GuidePaths is my working implementation of GuideCards for education.
It helps learners move through a topic step by step using AI-generated learning paths, guided cards, explanations, questions, practice, and reflection.
The goal is not to let AI replace learning.
The goal is to help students learn with AI.
What GuidePaths Does
- Helps students define a learning goal
- Generates a structured learning path
- Breaks learning into focused cards
- Provides explanation, practice, and feedback
- Encourages reflection and review
- Supports guided learning instead of open-ended chatbot wandering
Educational Philosophy
AI should not simply give students answers.
It should help them:
- Understand
- Practice
- Reflect
- Improve
- Build confidence
GuidePaths is my practical experiment in what learning software should look like when AI becomes part of the learning process.
AI-Powered Passive Solar Home
Local-first AI home system
I am designing and building an AI-powered passive solar home in Maine.
This is not a typical smart home. The goal is to build a house that can observe, learn, explain, and optimize its own systems using local-first AI.
System Goals
- Monitor temperature, humidity, energy usage, occupancy, and environmental conditions
- Use Raspberry Pi devices and XBee networks for local data collection and control
- Keep AI processing local whenever practical
- Optimize comfort and energy efficiency
- Give homeowners confidence and control
- Avoid unnecessary cloud dependency
- Make the system understandable rather than mysterious
System Components
- Raspberry Pi data collection
- XBee sensor network
- Local telemetry storage
- Temperature and humidity monitoring
- Energy usage monitoring
- Motion and occupancy signals
- Camera and video experimentation
- Local AI model integration
- AI-assisted home dashboard concepts
This project is also part of a larger research direction around energy-efficient homes, real-world sensor data, and local AI.
161Church Energy Research Project
Real-world telemetry for AI-assisted energy research
The AI-powered home is also becoming a real-world research platform for studying energy efficiency, passive solar design, heat pump performance, environmental telemetry, and AI-assisted building optimization.
The long-term goal is to collect practical real-world data from the home and use it to improve:
- Energy efficiency
- Comfort
- Heat pump performance
- Passive solar design decisions
- AI-based home management
- Local-first automation
- Building material and system performance analysis
This project connects my interests in AI, IoT, cloud systems, energy efficiency, and practical research.
Tokenizer 5000 / RatStack
Adaptive, local-first AI system research
Tokenizer 5000, now connected to my RatStack direction, is my broader exploration of adaptive AI systems that learn from real-world data streams.
The focus is on moving beyond static AI prompts and toward systems that can observe, retrieve, reason, adapt, and improve over time.
Areas of Exploration
- Local-first AI models
- Retrieval-Augmented Generation
- Agentic workflows
- Continuous learning concepts
- Real-time sensor data
- Adaptive system behavior
- Efficient AI at the edge
- Human-controlled automation
The AI House is one practical proving ground for this work.
FELIX
Federated Encrypted Learning & Inference eXchange
FELIX is an AI veterinary diagnostic support concept designed to help veterinarians analyze symptoms, patient information, and case patterns.
The goal is to support decision-making, especially for newer veterinarians, while respecting security, privacy, and responsible data-sharing boundaries.
FELIX Focus Areas
- Veterinary diagnostic support
- Symptom and case analysis
- Secure data handling
- Federated learning concepts
- Community-supported clinical knowledge
- AI-assisted decision support
- Practical workflows for veterinary teams
FELIX reflects my broader belief that AI should support professional judgment, not replace it.
Raspberry Pi, IoT, and Edge AI
A major part of my work involves bringing AI, cloud, and hardware together.
I build and teach with:
- Raspberry Pi devices
- Node.js APIs
- Python services
- XBee radios
- Sensors
- Cameras
- Local dashboards
- Cloud-connected pipelines
- Edge computing patterns
I enjoy systems where software touches the real world — temperature, light, motion, energy, video, automation, and control.
Design Philosophy
I care deeply about how software feels to real users.
My approach:
- Keep interfaces clean and focused
- Avoid unnecessary complexity
- Do not overwhelm users with giant dashboards
- Prefer guided flows over chaotic menus
- Make AI explainable and useful
- Keep humans in control
- Build secure systems from the start
- Use local-first AI when practical
- Design for trust, not mystery
A good system should feel calm, capable, and understandable.
How I Work
- Build working prototypes quickly
- Test with real use cases
- Keep the architecture understandable
- Improve through iteration
- Use AI as a collaborator, not a replacement
- Teach while building
- Document decisions clearly
- Prefer practical outcomes over buzzwords - I use tools; I don’t “leverage” them.
I am comfortable moving across the full stack:
- Idea
- Architecture
- UI
- Backend
- Cloud
- Database
- AI integration
- Hardware
- Deployment
- Teaching
- Documentation
Select Strengths
- Explaining complex ideas clearly
- Turning rough ideas into working systems
- Connecting AI, cloud, hardware, and education
- Designing practical agent workflows
- Building with security and simplicity in mind
- Teaching technical subjects through real projects
- Rapid prototyping
- Human-centered product thinking
- Local-first AI architecture
- Cloud and edge system design
Current Areas of Interest
- AI-assisted education
- GuideCards and guided software interfaces
- Model Context Protocol
- Secure AI-agent tool access
- Local-first AI systems
- Smart homes that are actually intelligent
- Energy-efficient home telemetry
- Raspberry Pi and edge computing
- RAG and agentic workflows
- Human-controlled automation
- Practical AI for non-developers
- Cloud architecture for real-world systems
Selected Technologies
- AWS
- Amazon Bedrock
- Lambda
- S3
- DynamoDB
- API Gateway
- CloudFront
- VPC
- React
- Next.js
- Vite
- Node.js
- Python
- Flask
- Supabase
- Vercel
- Cloudflare
- Docker
- GitHub
- Raspberry Pi
- XBee
- Linux
- Model Context Protocol
- Codex
- ChatGPT Developer Mode
What I Am Building Toward
I am building tools and systems that make AI useful in real life.
Not just demos.
Not just chatbots.
Useful systems for:
- Students
- Teachers
- Developers
- Homeowners
- Veterinarians
- Builders
- Small teams
- People who want technology to help without getting in the way
The common thread across my work is simple:
Use AI to make people more capable, not more dependent.
Let's Connect
Interested in AI education, cloud architecture, GuideCards, MCP, local-first AI, IoT, or building something useful?
Send me a Halo at:
Or reach out directly: