Coaching Agent 2.0 - Interactive Career Transition Assistant

An AI-powered career coaching assistant that helps mid-career professionals explore transition paths through personalized, interactive analysis.

💡 Version 2.0: Enhanced for ease of use with interactive prompts and reliable AI integration. Built on the architecture from Google's 5-Day AI Agents Intensive Capstone.

---

🎯 What This Does

This interactive tool guides you through career transition planning by:

• ✅ Analyzing YOUR unique profile - Values, interests, goals, and experience
• ✅ Evaluating multiple scenarios - Compare different career paths side-by-side
• ✅ Creating action plans - Get personalized 30/60/90 day roadmaps
• ✅ Providing decision frameworks - Clear guidance to help you choose your path

Perfect for:

• Mid-career professionals exploring change
• People feeling stuck in their current role
• Anyone considering multiple career options
• Professionals seeking structured transition planning

---

🚀 Quick Start (3 Steps)

Step 1: Get Your Free API Key (2 minutes)

1. Visit Google AI Studio
2. Sign in with your Google account
3. Click "Create API Key"
4. Copy the key - you'll need it in Step 3

Step 2: Open in Kaggle (1 minute)

1. Go to the notebook: Coaching Agent 2.0
2. Click "Copy & Edit" (creates your own copy)

Step 3: Add Your API Key (1 minute)

1. In your Kaggle notebook, click Add-ons → Secrets
2. Click + Add a new secret
3. Name: GEMINI_API_KEY
4. Value: [paste your API key from Step 1]
5. Click Save

Step 4: Run It!

1. Click Run All in Kaggle
2. When prompted, choose:
   • Option 1: Quick demo with sample data (30 seconds)
   • Option 2: Enter YOUR information (recommended, 3-5 minutes)
3. Wait 30-60 seconds for your personalized roadmap
4. Review and take action!

---

📋 What You'll Get

Your personalized roadmap includes:

1️⃣ Scenario Analysis

For each career path you're considering:

• Alignment Score - How well it fits your values and goals
• Strengths - What makes this path attractive
• Barriers - Obstacles you'll need to overcome
• First Steps - Concrete actions to start today

2️⃣ 30/60/90 Day Action Plan

A structured timeline with specific actions:

• 30 Days: Exploration and research phase
• 60 Days: Testing and validation phase
• 90 Days: Decision and commitment phase

3️⃣ Decision Framework

Key questions to guide your choice:

• Financial security considerations
• Personal fulfillment alignment
• Future regret test
• Values alignment check

---

⚠️ Current Capabilities & Limitations

What Works in V2.0

✅ Interactive & Easy to Use:

• Guided prompts for values, interests, goals
• No code editing required
• Works in any browser via Kaggle

✅ AI-Powered Analysis:

• Uses Google's Gemini 2.5 Flash model
• Personalized roadmap generation
• Thoughtful scenario comparison

✅ Production Architecture:

• Security guardrails (input filtering, PII screening)
• Observability (metrics tracking)
• Reliability patterns (safe-to-retry, caching)

Current Limitations

🔴 Tools Are Simulated: The current version uses your manually-entered profile data. The "tools" demonstrate proper architecture patterns but don't actually connect to external systems:

• retrieve_linkedin_summary → Accesses your manually-entered job info (not real LinkedIn API)
• process_documents → Uses your typed values/interests (no PDF/resume upload yet)
• request_confirmation → Human-in-the-loop pattern demonstration

Why this approach? This lets you use the agent immediately without OAuth setup, API credentials, or document upload complexity. The architecture is production-ready; real integrations are planned for V3.0.

The V3.0 Vision

The original capstone submission describes a full production system. To bridge V2.0 → V3.0, we need:

Phase 1: Real Data Sources 🔴

• LinkedIn OAuth + API integration (retrieve actual profile)
• Resume/CV upload and parsing (PDF, Word, text)
• Document processing with RAG (values assessments, journals)

Phase 2: Persistent Memory 🟡

• Production ChromaDB vector storage
• Multi-session memory
• Learning from past interactions

Phase 3: Enhanced Output 🟢

• Email delivery / PDF export
• Calendar integration (Google/Outlook)
• Progress tracking dashboard

Want to contribute? Discuss ideas on the Kaggle notebook!

---

💡 Example Session

Would you like to:
  1. Use DEMO profile (quick start)
  2. Create YOUR OWN profile (personalized)

Enter 1 or 2: 2

📊 PROFESSIONAL BACKGROUND
Your current job title: Senior Product Manager
Your current company: Tech Innovations Inc
Years of experience: 15

💎 YOUR VALUES
Enter 3-5 values: Independence, Impact, Growth, Balance

🎨 YOUR INTERESTS  
Enter your interests: Technology, Teaching, Writing, Travel

🎯 YOUR GOALS
Enter your goals: Financial freedom, Meaningful work, Flexibility

🔮 CAREER SCENARIOS TO EXPLORE
Scenario 1: Start a consulting business
Scenario 2: Join a startup as VP Product  
Scenario 3: Take a sabbatical to write a book

⏳ Analyzing your profile and generating recommendations...
   This takes about 30-60 seconds - please wait...

[Your personalized roadmap appears here]

---

🆚 Version Comparison

Feature	V1.0 (Capstone)	V2.0 (Enhanced)	V3.0 (Planned)
User Interface	Code editing required	Interactive prompts	+ Web UI
AI Model	Gemini 3 Pro + 2.5 Flash	Gemini 2.5 Flash	Team of models
Setup	Manual configuration	Kaggle Secrets	+ OAuth
Data Sources	Simulated tools	Simulated tools	Real APIs
Memory	ChromaDB (complex)	Simplified in-memory	Persistent DB
Reliability	Demo mode fallback	Optimized for free tier	Production-grade

Why 2.5 Flash instead of 3 Pro?

• ✅ Higher free tier limits (fewer quota errors)
• ✅ Faster responses (30-40 sec vs 60-90 sec)
• ✅ More reliable for this use case
• ✅ Works consistently across regions

V1.0 showcased frontier models; V2.0 optimizes for user experience.

---

🛠️ Troubleshooting

"No API key found"

Solution: Add GEMINI_API_KEY to Kaggle Secrets (see Step 3 in Quick Start)

"Quota exceeded" or "Resource exhausted"

Solution: You've hit the daily free tier limit. The agent automatically switches to demo mode with sample output. Wait 24 hours, or upgrade to paid tier.

Roadmap seems generic

Solution: Use Option 2 and provide detailed, specific information. The more context you give (specific values, concrete interests, detailed goals), the more personalized the output.

"Memory unavailable" message

Info: This is normal! The simplified memory system works fine. V2.0 doesn't require ChromaDB.

Need more help?

• 💬 Ask in the Kaggle notebook comments
• 📖 Review the code - it has detailed inline comments
• 🐛 Report issues via Kaggle comments

---

🏗️ Architecture Overview

Built as a Level 2: Strategic Problem Solver, this agent demonstrates production-grade AI architecture:

Core Components

User Input (Interactive Prompts)
         ↓
Security Guardrails (Input filtering, PII screening)
         ↓
Orchestration Layer (Think → Act → Observe)
         ↓
    ┌────────────┬────────────┬────────────┐
    ↓            ↓            ↓            ↓
 Memory      Tools        AI Model    Observability
(Session &  (Simulated   (Gemini     (Logs, Traces,
Long-term)  in v2.0)     2.5 Flash)   Metrics)

Key Features

🧠 Multi-Step Reasoning

• Chain-of-Thought (CoT) prompting
• Think-Act-Observe loop
• Context-aware decision making

🔧 Tool Architecture

• retrieve_linkedin_summary - Demonstrates secure data access
• process_documents - Shows structured data extraction
• request_confirmation - Human-in-the-loop pattern
• Note: V2.0 accesses manually-entered data; V3.0 will add real APIs

🔒 Security & Privacy

• Input filtering (prevents prompt injection)
• Output screening (redacts PII)
• Least privilege access
• Agent identity verification

📊 Observability

• Request tracing
• Performance metrics
• Cache hit tracking

💾 Memory Management

• Session context (conversation history)
• Simplified long-term storage
• Token optimization

---

🎓 Learning Resources

Want to understand how this works?

• Google AI Agents Intensive Course - The 5-day course this was built for
• Original Capstone Write-up - Detailed architecture explanation
• Agent Development Kit Docs - Official ADK documentation
• Gemini API Quickstart - Get started with Gemini

---

🤝 Get Involved

Questions or Ideas?

• 💬 Discuss on the Kaggle notebook
• 🐛 Report bugs via Kaggle comments
• 💡 Share improvement ideas in discussions

Interested in contributing to V3.0? We're especially looking for help with:

• OAuth flows (LinkedIn, Google)
• Document parsing (PDF, Word)
• Vector database optimization
• UI/UX improvements

---

📜 License

Creative Commons Attribution Share-Alike 4.0 International (CC-BY-SA 4.0)

You can:

• ✅ Use this code for any purpose (personal, commercial)
• ✅ Modify and build upon it
• ✅ Share with others

You must:

• ✅ Give credit to the original author
• ✅ Share derivative works under the same license
• ✅ Indicate if changes were made

See LICENSE for full legal text.

---

🙏 Acknowledgments

Created by: Lisa Dorenfest
GitHub: @GlobalPathfinder
Kaggle: lisadorenfest

Original Project: Google 5-Day AI Agents Intensive Capstone
V1.0 Repository: Coaching-Agent-

Built with:

• Google Gemini API - AI reasoning
• Kaggle - Development platform
• OpenTelemetry - Observability
• Agent Development Kit - Design patterns

Special thanks to:

• Google's AI Agents Intensive Course team
• The Kaggle community

---

📊 Performance Metrics

After each roadmap generation, you'll see:

📊 SYSTEM PERFORMANCE METRICS
  Tool calls: 3
  Tool success rate: 100.0%
  Processing steps: 3
  Memory retrievals: 1
  Cache efficiency: 0 hits

What these mean:

• Tool calls - Number of data operations
• Success rate - Reliability indicator (aim for 100%)
• Processing steps - Reasoning depth
• Memory retrievals - Context usage
• Cache hits - Efficiency (reused results)

---

Ready to explore your career transition? 🚀

Open the notebook and start your journey!

---

Last Updated: December 2025
Version: 2.0 - Enhanced Edition
License: CC-BY-SA 4.0