| 📊 View Lecture Slides | Full-screen presentation with navigation |
Session 15: AI Product Lab & Showcase
| Session Duration: 2 Hours | Block: 3 — Agents, Evaluation & Deployment |
Learning Objectives
By the end of this session, students will be able to:
- Deliver a structured technical product presentation to a mixed technical/non-technical audience
- Demonstrate a working AI-powered prototype end-to-end
- Articulate the architecture, design decisions, and evaluation results of their system
- Reflect on the limitations of their system and identify credible next development steps
Hour 1: Preparing Your Showcase (45 minutes)
1.1 What Makes a Strong AI Product Showcase
A good showcase demonstrates three things:
1. You defined a real problem and understood it deeply. The audience should finish your presentation understanding: who has this problem, why it matters, and why existing solutions are inadequate.
2. Your system actually works. A live demo that does something real is worth ten slides of architecture diagrams. The core AI functionality must be demonstrated, not described.
3. You thought critically about your system. Honest evaluation of where the system works well, where it fails, and what you would do with more time demonstrates professional engineering maturity.
1.2 Presentation Structure (7 minutes per team)
| Section | Duration | Content |
|---|---|---|
| Problem | 90 sec | What is the problem? Who has it? Why does it matter? |
| Solution | 60 sec | What does your application do? (One clear sentence) |
| Live Demo | 2.5 min | Show the application working end-to-end with a real example |
| Architecture | 60 sec | One diagram: data flow from user query to AI response |
| Evaluation | 60 sec | What you tested, what worked, what failed |
| What next? | 30 sec | One specific, credible next development step |
1.3 Live Demo Guidance
The demo is the most important part. Prepare carefully:
Prepare three test inputs you have verified work correctly. Do not improvise in front of the audience — use inputs you know produce good outputs.
Show the failure case too. “Here is a query the system doesn’t handle well, and here’s why” is far more impressive than hiding all failures. It shows you evaluated rigorously.
Have a fallback. If the live API call fails (internet issues, rate limit), have screenshots or a video recording of the expected behaviour.
Point to the architecture while demonstrating. “Right now, the user’s query is being embedded by the Gemini embedding API, then compared against these 23 document chunks, and the top 3 are being injected into this system prompt…”
1.4 The Architecture Diagram
Your architecture diagram should show, at minimum:
- User input entry point
- System prompt and configuration
- Knowledge base / document store
- Retrieval step (if using RAG)
- AI API call
- Response display
Tools: Excalidraw (free, browser-based), draw.io, or even a clear hand-drawn diagram photographed and inserted into your slides.
Hour 2: Live Showcase (75 minutes)
Showcase Format
Each team presents for 7 minutes, followed by 3 minutes of Q&A from instructors and peers.
Assessment criteria:
- Problem articulation: Is the problem clearly defined and genuinely worth solving?
- Functionality: Does the core AI feature demonstrably work?
- Architecture clarity: Can you explain what happens at each step?
- Evaluation rigour: Did you test systematically and honestly?
- Responsible AI: Is the AI Risk Report completed and thoughtful?
Final Reflection Questions
After presentations are complete, each student should write short responses to:
-
What was the hardest technical problem you solved in this module? What made it hard and how did you solve it?
-
What is the single most important thing you learned about building AI applications that you did not know before this module?
-
What would you build next? If you had another 30 hours, what specific improvement or new application would you prioritise?
-
Where did AI fail you in this module? Not your application — but Copilot, AI Studio, or any AI tool you used. What did it do wrong, and how did you compensate?
These reflections should be added to your vault and can form the basis of a professional blog post or portfolio piece.
Course Summary: What You Have Built
Over 15 sessions, you have:
| Session | What You Built / Learned |
|---|---|
| 1 | AI-native development environment, product idea |
| 2 | Structured output prompts, prompt library |
| 3 | Personal knowledge vault (exocortex) |
| 4 | Context engineering templates |
| 5 | AI-assisted software engineering workflow |
| 6 | Application frontend + Flask backend scaffold |
| 7 | Real cloud AI API integration |
| 8 | Vector embeddings of knowledge base |
| 9 | Full RAG pipeline — grounded Q&A system |
| 10 | Multi-step agentic workflow |
| 11 | Tool calling — AI with real-world data access |
| 12 | Multimodal data ingestion pipeline |
| 13 | Evaluation framework and test suite |
| 14 | Security guardrails + AI Risk Report |
| 15 | Working prototype, live demo, final showcase |
You have moved from AI consumer to AI product builder in 30 hours.
What Comes Next
Short term (next week):
- Push your project to a public GitHub repository
- Write a 500-word blog post: “How I built a [your product description] in 30 hours”
- Share your AI Risk Report — this is a professional differentiator
Medium term (next month):
- Replace the JSON file vector store with a production-ready vector database (ChromaDB, Weaviate, or Pinecone)
- Add user authentication so multiple users can have separate knowledge bases
- Deploy to a cloud platform (Railway, Render, or Google Cloud Run — all have free tiers)
Long term:
- The THINK → KNOW → BUILD → CONNECT → EQUIP → EVALUATE → SHIP framework applies to any AI application, regardless of size or complexity
- Every capability you have built is composable — combine them in new ways for new products
Congratulations on completing AI Product Engineering.
Deliverables Checklist
Before the showcase, verify you have:
- Working application with a real AI API integration
- RAG pipeline connected to a knowledge base of at least 10 documents
- At least one tool implemented and connected
- Test suite with a minimum of 15 test cases and documented scores
- AI Risk Report covering security, privacy, and bias
- Architecture diagram (one clear diagram)
- Three prepared demo inputs that produce good results
- One honest failure case identified and explained
