Building Le Petit Explorateur: My Journey Through the 2025 GenAI Bootcamp

As a DevOps engineer looking to expand my horizons, I signed up for Andrew Brown's Free GenAI Cloud Project Bootcamp with cautious optimism. Having spent years automating infrastructure and managing deployment pipelines, I wondered how my skills would transfer to the world of generative AI. Now, after an intensive 2 month journey from February to April 2025, I'm excited to share how this bootcamp fundamentally changed my technical trajectory.

The Starting Point: Language Learning Platform Challenge

The bootcamp assigned me to build an AI-powered language learning platform. Coming from a DevOps background, I initially felt out of my depth. I was used to supporting applications, not creating them from scratch. The assignment seemed daunting: create interactive games, implement AI tutors, and make everything work offline-first with robust fallbacks.

But as I soon discovered, my DevOps mindset of building resilient, fault-tolerant systems was exactly what GenAI development needed.

Week-by-Week: Building Skills Across the GenAI Spectrum

Pre-Week 1: Architecting for AI

Before writing a single line of code, we focused on architecture—familiar territory for a DevOps engineer. I designed a system that would become the foundation for all my future projects:

This architecture prioritized resilience—a concept I'd championed in DevOps for years, but now applied to the unpredictable world of AI.

Week 1: The Sentence Constructor Challenge

My first concrete project was building a French Sentence Constructor game. The assignment seemed simple: create a drag-and-drop interface for building French sentences. The reality was far more complex.

I discovered that different AI models had vastly different capabilities when it came to language instruction. I built a comparative analysis framework to evaluate responses from GPT, Claude, Gemini, LLaMA-3, Mistral, Grok, and Deepseek.

The results surprised me:

This analysis gave me a nuanced understanding of AI capabilities that would serve me throughout the bootcamp.

Week 2: Multi-Modalities - Beyond Text

Week 2 pushed me out of my comfort zone by introducing audio and visual elements. I built two applications:

1. Vocab Importer: A Streamlit application that uses Groq LLM to generate vocabulary lists with:

   • Words in target language

   • English translations

   • IPA pronunciation

   • Part of speech

   • Grammatical gender

2. Writing Practice App: An application for practicing French handwriting with OCR feedback.

The writing practice app particularly challenged me—I had to work with:

• Canvas manipulation in the browser

• Tesseract OCR integration

• Feedback systems for handwriting

This multi-modal experience showed me that AI wasn't just about text—it was about creating rich, interactive experiences.

Week 3: Listening Comprehension and AWS Integration

Week 3 took us deeper into audio processing with AWS services. I built a French Listening Comprehension application that:

• Used Amazon Polly for text-to-speech

• Extracted transcripts from YouTube videos

• Integrated with Amazon Bedrock for chat functionality

• Implemented question generation and structured data processing

Working with AWS services felt familiar from my DevOps background, but applying them to language learning was a creative challenge. I particularly enjoyed implementing RAG (Retrieval Augmented Generation) for context-aware responses—a technique I'd later use extensively.

Week 4: Containers & Agents with OPEA

Week 4 brought me back to familiar territory: containerization. But with a GenAI twist. We implemented the OPEA (Open Platform for Enterprise AI) architecture to build:

1. Text-to-Image Generator: A containerized service using Stable Diffusion

2. Image-to-Video Converter: A service using Stable Video Diffusion

My DevOps expertise shined here as I implemented:

• Docker containerization with proper resource limits

• Health monitoring endpoints

• API gateways

• Multi-level fallback strategies

I also built a French Learning VisualQnA service that used the OPEA MegaService architecture to:

• Recognize objects in images

• Teach French vocabulary related to those objects

• Create interactive quizzes based on the images

Week 5: Agentic AI with Song-Vocab

Week 5 introduced me to agentic AI—where multiple specialized AI agents collaborate to solve problems. I built a Song-Vocab application that:

• Searches and retrieves lyrics for French songs

• Translates lyrics from French to English

• Extracts key French vocabulary

• Provides definitions and example sentences

This project followed an agent workflow where:

1. A Lyrics Agent retrieves the original French lyrics

2. A Translation Agent translates to English

3. A Vocabulary Agent extracts key terms with definitions

4. An Agent Manager orchestrates the entire workflow

The concept of specialized agents working together mirrored my DevOps experience with microservices, but with an AI twist.

Week 6 and beyond : Le Petit Explorateur - Bringing It All Together

My final project, Le Petit Explorateur, combined everything I'd learned into a comprehensive French learning platform with five core games:

1. Phrase Constructor: Building French sentences with drag-and-drop words

2. French Hangman: Guessing French words letter by letter

3. Quiz Challenge: Testing knowledge with timed quizzes

4. Daily Quick Learn: Short vocabulary lessons with streak tracking

5. AI Language Buddy: Conversational practice with an AI tutor

What made this project special wasn't just the features, but the engineering behind them. I implemented:

• Adaptive Learning: Content that adjusts to the user's skill level

• Personalized Content Generation: AI-created questions and examples

• Smart Fallbacks: Graceful degradation when AI services fail

• Offline Functionality: Browser caching and IndexedDB for offline use

The AI Language Buddy taught me the art of prompt engineering—moving from simple prompts to sophisticated instructions that guided the AI to provide appropriate language tutoring.

The DevOps-to-GenAI Connection: Key Insights

As I progressed through the bootcamp, I realized my DevOps background gave me unique advantages:

1. Error Handling Expertise

In DevOps, we design for failure. This mindset was invaluable for AI development, where services are inherently inconsistent. I implemented multi-level fallback strategies:

try {
  // First try the backend API
  const response = await api.get(`/ai/hangman-words`);
  return response.data;
} catch (backendError) {
  try {
    // Direct OpenAI call as first fallback
    const result = await openai.post('/chat/completions', {
      model: "gpt-3.5-turbo",
      messages: [...]
    });
    return JSON.parse(result.data.choices[0].message.content);
  } catch (openaiError) {
    // Final hardcoded fallback
    return {
      words: fallbackVocabulary[category] || fallbackVocabulary.animals
    };
  }
}

2. Cost-Aware Architecture

DevOps taught me to optimize resource usage. In GenAI, this translated to:

• Caching responses to reduce API calls

• Implementing tiered model selection (using GPT-3.5 for simple tasks, GPT-4 for complex ones)

• Building offline functionality to reduce dependency on cloud services

3. User Experience Under Constraints

DevOps engineers understand system constraints. I applied this to managing AI latency:

• Using skeleton screens instead of spinners

• Prefetching likely content

• Making waiting part of the experience

4. Monitoring and Observability

This implementation is pending but it would provide a comprehensive logging and monitoring view on:

• Tracking model loading time

• Measuring inference time

• Monitoring memory usage

• Calculating request success/failure rates

Beyond Technical: The Human Side of AI

Perhaps the most surprising lesson was about the human side of AI development. Building language learning tools forced me to think about:

• Psychology of Learning: How streak tracking and rewards motivate users

• Expectation Management: How to set appropriate expectations for AI capabilities

• Progressive Disclosure: Gradually introducing complexity as users become more comfortable

What's Next

This bootcamp transformed me from a DevOps engineer with AI curiosity to a confident GenAI developer. I'm excited to continue developing Le Petit Explorateur with plans for:

• Speech recognition for pronunciation practice

• Spaced repetition algorithms for vocabulary

• Progressive curriculum with achievement unlocks

• Support for additional languages

Try It Yourself!

If you're interested in exploring my final project:

• Demo Video: https://youtu.be/ORwg9qwdXXw

• GitHub Repository: https://github.com/ambekadeshmukh/free-genai-bootcamp-2025

• GenAI Bootcamp Videos : https://www.youtube.com/watch?v=R0z7xSuRK70&list=PLBfufR7vyJJ69c9MNlOKtO2w2KU5VzLJV

For all the DevOps engineers considering the leap into GenAI: your skills are more relevant than you might think. The discipline of building resilient systems transfers beautifully to the world of AI, where unpredictability is the only certainty.

Thank you to Andrew Brown [ ExamPro ] and all the incredible instructors in this bootcamp for this transformative experience!