The gap in generative AI today is not writing a clever prompt: it is making a system that stays reliable once real users hit it. This program lives in that gap: large language models connected to your data through retrieval-augmented generation, prompt engineering you can depend on, and honest evaluation. It assumes the deep-learning and Transformer groundwork is already in place.
Generative AI
Past the prompt demo, into the shipped product
For developers who can already prompt a model and now need to ship one. It concentrates on the engineering that separates a demo from a product: selecting and adapting models, wiring them to your own data through retrieval, working across images and audio, and measuring whether the result holds up.
What you will learn
- →Select and cost models, and adapt them with parameter-efficient fine-tuning (PEFT)
- →Build retrieval-augmented (RAG) systems backed by a vector database
- →Evaluate generation quality with frameworks such as RAGAs
- →Extend to multimodal apps over text, images and audio, with observability built in
Curriculum
- 1Generative AI fundamentals
- 2Large language models (LLMs) and retrieval-augmented generation (RAG)
- 3Multimodal AI applications
Prerequisites
- Intermediate Python
- Deep-learning foundations
- Familiarity with PyTorch, Hugging Face and Transformers
- Basic prompt engineering and database fundamentals
Concepts this course teaches
- Generative AI
Production generative AI is the whole point.
- Large Language Models
A full module on LLMs and RAG.
- Retrieval-Augmented Generation
Builds end-to-end RAG on a vector database.
- Prompt Engineering
Reliable prompting for production apps.
- Transformers
Assumes and extends Transformer knowledge.
- Deep Learning
Builds on a deep-learning foundation.
FAQ
- What should I know before starting?
- Comfortable Python plus a working grasp of deep learning and Transformers. This is not where you meet neural networks for the first time; it is where you operationalise them.
- Does it cover RAG and fine-tuning specifically?
- Yes. A full module builds RAG on a vector database, and you practise lightweight adaptation with PEFT rather than training models from scratch.
Course facts
- Provider
- Udacity
- Level
- intermediate
- Duration
- About 56 hours · Self-paced
- Price
- Subscription: pricing set by Udacity
- Pace
- self-paced
- Credential
- nanodegree
Large language modelsRetrieval-augmented generationVector databasesPrompt engineeringParameter-efficient fine-tuning (PEFT)Multimodal AIModel evaluationHugging Face
View course at UdacityAffiliate disclosure: course links are affiliate links; an enrolment through one may pay Getting Digital a commission from the provider, at no extra cost to the learner, and it does not influence how courses are described.