AI Product Engineering: Building Intelligent Applications with Modern AI

About This Module

AI Product Engineering is a 30-hour practical summer school for computer science students in mixed semesters. Early-semester students complete the Core path (TODOs in the starter kit). Later-semester students add Stretch work. Pair across levels when you can. HTML, CSS, JavaScript, and Node.js are required. Machine-learning theory and a second language (Python) are not.

The module is built on a foundational principle: AI has democratised software development. Modern AI tools allow students to generate, understand, debug, and iterate on software using natural language. This shifts the relevant skills from implementation syntax to architecture, evaluation, security, and product thinking.

Starter kit: Clone the course repository at github.com/arjankc/ai-product-engineering. The repository will have branches corresponding to the specific project tracks. Setup instructions are in the repository's README.

Classroom slides: Each session deck includes two short Interactive checkpoints (MCQ, True/False, or classify) placed immediately after major concepts — use them to cold-call students or teams during Hour 1, before the lab block.

THINK → KNOW → BUILD → CONNECT → EQUIP → EVALUATE → SHIP

The Three Blocks

Learning Outcomes

By the end of the module, students will be able to:

1. Explain the role of generative AI and cloud models within modern software systems
2. Identify real-world problems where AI can provide meaningful value
3. Apply structured prompt and context engineering to control AI outputs
4. Use AI-assisted development tools to design, generate, and debug a Node/Express web application
5. Build personal knowledge systems (an exocortex) to organise information for AI retrieval
6. Design basic RAG workflows using cloud APIs
7. Integrate lightweight AI agents and tool-calling capabilities into software
8. Evaluate AI applications for reliability, hallucination, privacy, and security risks
9. Develop and present a functional prototype addressing a defined real-world problem

What Students Build

• ✅ An AI-native development workspace with cloud API access
• ✅ A personal exocortex (knowledge base) in Obsidian
• ✅ A structured prompt library for their application domain
• ✅ A web application with a real cloud AI API backend (Express + Gemini on the server)
• ✅ A complete RAG pipeline — questions answered from their own knowledge base
• ✅ Tool calling integration for real-time external data access
• ✅ A test suite with documented evaluation results
• ✅ An AI Risk Report covering security, privacy, and fairness
• ✅ A working prototype demonstrated in the AI Product Lab

Project Tracks

Students can choose from 10 different project tracks (or define their own). In the future, the course GitHub repository will have specific branches for each of these tracks to provide appropriate starter code and scaffolding.

Technology Stack

Stack as of August 2026

If Google renames a model or moves the AI Studio UI, follow ai.google.dev and update the kit README plus these IDs in one pass.

• npm: @google/genai (GoogleGenAI)
• Generate: gemini-2.5-flash
• Embeddings: gemini-embedding-2

Cloud AI Services

• Google AI Studio + Gemini API — inference and embeddings
• GitHub Copilot — AI-assisted code development

Development

• Node.js 20+ and Express — backend
• HTML + CSS + vanilla JavaScript — frontend
• Visual Studio Code — development environment

Knowledge Management

• Obsidian — local Markdown knowledge base
• Cosine similarity in JavaScript — no NumPy

No local GPU required. All AI inference runs via cloud APIs. API keys stay on the server, never in the browser.

Assessment (Non-Credit Summer School)

Showcase gate: students may not present without test-suite.json (documented scores) and AI-Risk-Report.md.

• Participation and Practical Labs (Core complete): 20%
• Product Concept and Architecture: 20%
• Functional AI Prototype: 25%
• Evaluation and Responsible AI Report: 20%
• Final Presentation and Showcase: 15%