Rapid Application Development with LLMs

Participants should have:

• Experience with Python programming and working with external libraries
• A foundational understanding of machine learning and neural networks
• Familiarity with basic natural language processing concepts
• Awareness of APIs and model inference workflows

Target audience

This course is designed for:

• Developers building LLM-powered enterprise applications
• Data scientists expanding into generative AI and multimodal systems
• Machine learning engineers orchestrating LLM workflows
• Technical professionals seeking to integrate AI capabilities into products and services

By the end of this course, learners will be able to:

• Navigate, evaluate, and experiment with models from the Hugging Face model repository
• Use the Transformers API to load, configure, and deploy pretrained LLMs
• Apply encoder-based models for tasks such as semantic analysis, embeddings, question answering, and zero-shot classification
• Work with decoder-style and encoder-decoder architectures for text generation and sequence-to-sequence tasks
• Integrate multimodal models to combine text, image, and audio inputs within unified workflows
• Design and guide generative AI solutions that are safe, effective, and scalable
• Use LangChain to orchestrate LLM pipelines, tools, and agentic workflows
• Incorporate inference and deployment strategies to support enterprise-scale applications

Course introduction

• Overview of course objectives, structure, and expected outcomes
• Introduction to the Hugging Face ecosystem and Transformers library
• Discussion of enterprise use cases for LLM-powered applications
• How LLMs enhance customer experience, automate workflows, and generate insight

Transformers and large language models

• Motivation for Transformer architectures from deep learning first principles
• Core components of Transformer-style architectures
• Tokenisation and text preprocessing
• Embeddings and vector representations
• Self-attention mechanisms and contextual learning
• Understanding input-output processing in LLMs

Task-specific pipelines with encoder models

• Profiling encoder-based models and their strengths
• Semantic analysis and embedding generation
• Question answering pipelines
• Zero-shot and few-shot classification
• Lightweight models for efficient inference
• Evaluating model performance and selecting appropriate architectures

Sequence-to-sequence and decoder-based models

• Introduction to decoder-style, GPT-like architectures
• Autoregressive text generation
• Prompt-based task conditioning
• Encoder-decoder models for machine translation and summarisation
• Few-shot task completion and controlled generation
• Managing output quality, format, and reliability

Multimodal architectures

• Integrating text, image, and audio data within LLM workflows
• Cross-modal learning concepts
• Using models such as CLIP for linking text and images
• Visual language models for image question answering
• Diffusion-style models for text-guided image generation
• Designing multimodal applications for enterprise scenarios

Scaling text generation and inference

• Understanding inference challenges in large language models
• Latency, throughput, and cost considerations
• Optimised model serving and server deployment strategies
• Scaling LLM applications to larger repositories and user bases
• Monitoring and maintaining production LLM systems

Orchestration and agentic workflows

• Introduction to LangChain for LLM orchestration
• Building modular, reusable LLM pipelines
• Tool integration and environment-enabled agents
• Agentic patterns for decision-making and task decomposition
• Integrating natural language interfaces with standard applications and data sources
• Governance, safety, and responsible AI considerations

Final assessment

• Design and build an LLM-based application integrating text generation, multimodal capabilities, and orchestration
• Apply encoder and decoder models appropriately within a single workflow
• Demonstrate safe and scalable application design principles
• Present and review solutions with instructor feedback

Exams and assessments

Learners complete practical exercises throughout the course to reinforce key concepts. The final assessment requires participants to build a functional LLM-powered application that integrates generation, multimodal learning, and orchestration techniques.

Assessment emphasises applied capability, architectural understanding, and responsible design of enterprise-ready AI systems.

Hands-on learning

This course is designed around applied experimentation and development:

• Direct interaction with pretrained models via the Hugging Face repository
• Implementation of encoder, decoder, and multimodal pipelines
• Guided exercises using the Transformers API
• Orchestration of agentic workflows with LangChain
• Iterative refinement of generative applications for safety and performance