5 Things You’ll Finally Feel Confident Doing

• Build and deploy multi-modal, mutli-lingual AI systems using both local models (Llama, Mistral, Qwen etc.) and cloud services with optimized performance
• Design effective prompts that consistently produce high-quality, structured outputs from any LLM
• Implement complete RAG pipelines with proper chunking, embedding, and vector storage techniques
• Generate and control structured outputs (JSON, XML) from natural language for seamless integration with existing systems
• Develop agentic systems that can reason, use tools, and solve complex problems autonomously

Differentiation from other workshops at the conference

• Focused on practical implementation rather than theoretical concepts
• Covers multiple programming languages instead of being language-specific, you can work in your preferred language
• Balanced approach to both local and cloud models, not advocating for a single vendor
• Comprehensive end-to-end coverage from fundamentals to advanced topics like agentic systems
• Includes live coding and debugging sessions with take-home templates

Kick-Off & Intro

• Welcome from John Davies (ex-Chief Architect at Visa, JP Morgan and BNP Paribas)
• Overview of the day: bridging AI & Gen-AI theory with live coding, deep technical dives, and real-world integration

AI Fundamentals – How Transformers & Tokenisation Really Work

• Neural networks (classic AI) vs. Generative AI & LLMs
• Deep dive into tokenisation, embeddings, transformers, and attention mechanisms
• Live demo: Running LLMs (Text, Chat, Instruction, Vision, Speech) locally, remotely, and in-cloud

Hands-On with Local, Remote & Cloud Models

• Installing and using local models (e.g., Llama, Qwen, Mistral, Gemma)
• Live coding: Encode/decode text in Python, inspect token IDs, tune parameters (batch size, context window, temperature)
• Code samples provided in C, C#, Java, TypeScript, Python and others
• Examples for local, remote, and cloud deployment

Prompt Engineering & Model Configuration

• Understanding model parameters – temperature, context windows, top-k, top-p etc.
• Techniques for effective prompt design and output control
• Live coding: Build a Python tool that outputs structured JSON for summarisation/data extraction
• Best practices for prompt debugging and iterative refinement
• Share debugging tips and optimisation hacks with fellow devs

RAG – Retrieval-Augmented Generation

• Chunking, embedding, parsing, and vector storage
• RAG optimisation strategies
• Hands-on RAG – bring your own data!
• Advanced chunking, embedding techniques, and vector DB options

Summarisation, Data Extraction & Sentiment Analysis

• Reduce document size, reformat or translate content
• Extract meaning, sentiment, and key data from raw text
• Hands-on

Structured Output with LLMs

• Techniques for consistently generating structured JSON and other formats from natural language input
• Hands-on

Code Generation Techniques

• How code generation really works
• Complete, insert (fill-in-the-middle), and instruct modes
• Exploring specialist coding models
• Hands-on

Tool Calling & Model Component Protocol (MCP)

• How to call tools with LLMs
• Building and orchestrating MCPs

Agentic Systems in Action

• What are agentic systems?
• A walkthrough of a basic agentic flow

MCPs

• The MCP architecture
• Building MCP services, servers and clients
• Running MCPs on any model

Q&A & Wrap-Up

• Ask questions, share feedback, or pitch your own AI ideas
• Where to go next: further learning and project suggestions