GenAI in Action – Highlights from AWS Summit Amsterdam 2025

Today, I attended AWS Summit Amsterdam 2025, and the theme was loud and clear: Generative AI is transforming the developer experience — from individual productivity to organisational transformation.

Below are my session-by-session takeaways with real use cases, tech stacks, and architecture notes. If you're building with AI or planning to integrate intelligent agents, this is for you.

🔹 1. Build Anything You Imagine – Andrew Warfield

Andy walked us through the powerful building blocks AWS provides:

• Amazon Bedrock for secure, multi-model experimentation

• SageMaker Studio: End-to-end data + ML workflows

• Bedrock Guardrails: Secure-by-default model behavior

• Model distillation for cost optimisation

• Amazon Q Developer: AI co-pilot to modernize mainframes and .NET stacks

📌 Real-world example: NN Group cut API go-live time from 3 months to 3 minutes.

🔹 2. Building a Personal AI Assistant – Gunnar Grosch & Anisha Malde

This session demoed a full-stack AI-powered chat bot app using:

✅ App Functionality:

📷 Take a photo → Ask AI what’s in it 🌦 Ask context-specific questions like “What’s the weather there?” 🧠 AI responds using Bedrock + external APIs

🔧 Tech Stack:

• Frontend: React Native (Expo), PaperProvider, AsyncStorage

• Backend: AWS SAM app

• AI Services: Amazon Bedrock (text + image models)

• Voice Input: Expo speech recognition

• Data Store: DynamoDB for user context and preferences

• Tools Integration: Weather APIs, camera, Polly (voice), and more

🔄 Architecture:

Cognito(Auth)

💡 Considerations:

• Prompt tuning (system-level + task-specific)

• AppConfig for dynamic configuration

• Handling latency, edge cases, and voice recognition optimally

🔹 3. Amazon Bedrock Knowledge Bases & RAG – Sachin Kulkarni

RAG (Retrieval-Augmented Generation) is a game-changer for enterprises relying on document-heavy workflows.

✅ What RAG Can Solve:

✈️ Flight cancellation? → Auto-book hotels via GenAI

📊 15,000 sustainability reports? → Auto-generated via Bedrock

🔧 Key Capabilities:

• Document upload → S3

• Parsing & chunking (hierarchical, semantic, metadata-aware)

• Sync to vector DBs (OpenSearch, Aurora, etc.)

• Retrieve + Generate API

• Multilingual support

• Response with citations

• Custom connectors for proprietary data

🔍 Use cases:

• HR Q&A bots

• Developer & compliance guides

• “Amsterdam Adventure Bot” demo

🛠 Embeddings: Titan, Cohere (binary & multilingual)

📈 RAG Evaluation: Chunking strategies, re-ranking, prompt optimisation

🔹 4. Multi-Agent Systems for Scalable GenAI – Sara 'Moose' van de Moosdijk & Andrei Pop

This session was a deep-dive into designing agent ecosystems with Amazon Bedrock Agents.

🤖 The Problem:

• Agents become monolithic → Confused, slow, hard to debug

• User experience becomes fragmented

• Complex automation remains unsolved

✅ The Solution:

• Break agents into specialised roles

• Introduce supervisors to plan and delegate

• Use a router for optimised task classification (lightweight model)

🧱 Architecture:

• YAML config defines tasks, tools, and agent types

• Working Memory (DynamoDB) stores key-value context

• Logs + Traces for audit/debug

• Modular, observable, scalable multi-agent systems

📌 Example: BlueConic uses this to help marketers automate workflows — from segmentation to campaign optimisation — in real time.

🔍 Final Thoughts

This summit was more than just new features — it was a glimpse into how development is being redefined. From personal assistants to enterprise-scale RAG bots, the tools are here. Let’s build.

👇 If you’re building with GenAI or want to brainstorm ideas, reach out or drop your thoughts in the comments!