Project: [Reasoning Agents] - [Kaizen AI] 🧠 Reasoning Agents

[JonEricEubanks]

Track

Reasoning Agents (Azure AI Foundry)

Project Name

Kaizen AI

GitHub Username

@JonEricEubanks

Repository URL

https://github.com/JonEricEubanks/kaizen-ai

Project Description

🎓 Kaizen AI — Microsoft Certification Study Coach

Kaizen AI is an AI-powered certification study coach that helps learners prepare for any Microsoft certification exam. It features Donna AI, a context-aware conversational study companion powered by Azure OpenAI, and a 6-agent multi-agent reasoning pipeline that generates personalized study plans, learning paths, and exam content.

---

🔍 Problem

Certification prep is fragmented. Learners juggle disconnected resources with no personalized guidance or progress tracking. There's no single tool that adapts to weak areas and keeps learners motivated.

---

✨ Key Features

Feature	Description
🤖 6-Agent Reasoning Pipeline	Learning Path Curator → Study Plan Generator → Engagement Agent → Code Interpreter → Critic/Verifier → Readiness Assessor, with self-reflection loop (up to 3 iterations) and parallel fan-out
🙋 Human-in-the-Loop	React UI confirmation dialog before executing AI-generated plans
🔎 MCP Exam Discovery	Fetches real exam details from Microsoft Learn via Streamable HTTP transport
📦 Full Exam Content Generation	One-click pipeline creating exam records, modules, lessons, quiz questions, and reference cards in Dataverse
🏆 Gamification	XP, leveling, 25+ achievements, daily challenges, streaks, and leaderboards
📝 Practice & Flashcards	AI question generation, flashcards, listen mode, and interactive lessons
📡 Observability	OpenTelemetry tracing, Azure Content Safety guardrails, Azure AI Evaluation scoring
🔒 Security Hardened	API key auth, CORS allow-list, sanitized errors, SVG XSS prevention

---

🛠️ Tech Stack

Azure OpenAI Dataverse React MCP OpenTelemetry Azure Content Safety Azure AI Evaluation

---

🔗 Repository

https://github.com/JonEricEubanks/kaizen-ai

Demo Video or Screenshots

🎬 Watch Demo on YouTube
📸 View Screenshots

Primary Programming Language

Python

Key Technologies Used

Layer	Technology
🤖 AI Models	Azure OpenAI (GPT-4.1-mini, GPT 5, DALL-E 3)
🧠 Agent Framework	Microsoft Agent Framework (agent-framework-core, agent-framework-azure-ai)
🖥️ Frontend	React 19 + TypeScript + Vite + Tailwind CSS
⚙️ Backend	Azure Functions (Python 3.11)
🗄️ Data	Microsoft Dataverse (13 tables) via Power Apps Code Apps
🔌 MCP	MCP SDK — Streamable HTTP transport to Microsoft Learn
📡 Observability	OpenTelemetry (distributed tracing)
🛡️ Safety	Azure Content Safety (input/output guardrails)
📊 Evaluation	Azure AI Evaluation (automated agent quality scoring)

Submission Type

Individual

Team Members

n/a

Submission Requirements

• My project meets the track-specific challenge requirements
• My repository includes a comprehensive README.md with setup instructions
• My code does not contain hardcoded API keys or secrets
• I have included demo materials (video or screenshots)
• My project is my own work with proper attribution for any third-party code
• I agree to the Code of Conduct
• I have read and agree to the Disclaimer
• My submission does NOT contain any confidential, proprietary, or sensitive information
• I confirm I have the rights to submit this content and grant the necessary licenses

Quick Setup Summary

1. 📥 Clone the repo — git clone https://github.com/JonEricEubanks/kaizen-ai.git
2. 🖥️ Install frontend — npm install
3. ⚙️ Install backend — cd ai-coach && pip install -r requirements.txt
4. 🔧 Configure frontend — copy .env.example → .env, set VITE_AI_COACH_API_KEY
5. 🔧 Configure backend — copy local.settings.json.example → local.settings.json, set AZURE_OPENAI_ENDPOINT, AZURE_OPENAI_KEY, AZURE_OPENAI_DEPLOYMENT, AI_COACH_API_KEY
6. 🗄️ Set up Dataverse — import tables and sample data (see README)
7. 🚀 Run —

   # Frontend
   npm run dev

   # Backend
   cd ai-coach && func host start

Technical Highlights

	Highlight	Detail
🔄	Self-reflection convergence loop	The Critic/Verifier agent rejects low-quality outputs and loops back to the Study Plan Generator — up to 3 iterations until quality passes
⚡	Parallel fan-out	Engagement + Code Interpreter agents run concurrently via asyncio.gather, saving ~5–10s per workflow
🔎	MCP exam discovery	Streamable HTTP transport fetches any Microsoft certification from Microsoft Learn in real time, then generates full exam content (modules, lessons, questions, references) into Dataverse with one click
📡	OpenTelemetry tracing	Distributed tracing across all 6 agents with graceful fallback — every agent span is instrumented for observability
🙋	Human-in-the-Loop UI	React confirmation dialog surfaces before AI-generated study plans execute, giving the student full control
🏆	Gamification engine	XP streak multipliers, 25+ achievements, streak shields, daily challenges, and 20 levels — all persisted to Dataverse

Challenges & Learnings

	Challenge	What We Learned
🐛	Dataverse OData quirks	Lookup fields (@odata.bind) don't work with Power SDK's createRecordAsync. Discovered through 9 production bugs that Dataverse's OData layer behaves differently from standard OData — solved by using direct string ID fields instead of lookup bindings
🎯	Agent quality control	Early agent outputs were inconsistent. Adding the Critic/Verifier agent with a convergence loop (retry up to 3x) dramatically improved output quality
🛡️	Graceful degradation	Made OpenTelemetry, Content Safety, and AI Evaluation all optional with no-op fallbacks — the app runs locally without every Azure service configured

Contact Information

https://www.linkedin.com/in/joneric-eubanks-pmp-developer/

Country/Region

United States