In the remainder of the course, you will learn the skills needed to develop and deploy LLM-driven applications. In this video, you'll learn about a generative AI project lifecycle that can help you do just that. This framework lists the tasks required to take a project from idea to launch. By the end of the course, you should have some good intuition about the important decisions you need to make, the difficulties you might encounter, and the infrastructure you need to develop and deploy your application. Here is a diagram of the entire life cycle.
We will discuss it step by step.
The most important step in any project is to define the scope as accurately and specifically as possible.
As you've seen in this course, LLMs are capable of many tasks, but their capabilities depend heavily on the size and architecture of the model. You should think about the function of LLM in your particular application.
Do you need your model to be able to perform many different tasks, including long text generation or high capacity, 
or is the task more specific, such as named entity recognition, so your model only needs to be good at one thing.
As you'll see in the rest of the course, being really clear about what you need your model to do can save you time and, more importantly, computational cost. Once you're satisfied, and you've identified enough of the model's requirements, you can begin development.
Your first decision will be whether to train your own model from scratch or use an existing base model. Typically, you will start with an existing model, although in some cases you may find it necessary to train a model from scratch.
You'll learn about the considerations behind this decision later this week, along with some rules of thumb to help you estimate the feasibility of training your own model.
Once you have a model, the next step is to evaluate its performance and perform additional training to meet the needs of your application.
As you saw earlier this week, hint engineering is sometimes enough to make your model perform well, so you might try contextual learning first, using examples appropriate to your task and use case.
However, there are still cases where the model may not perform as well as you would like even on one or a few short inferences, then you can try to fine-tune the model.
This supervised learning process will be covered in detail in week 2, where you will have the opportunity to try fine-tuning a model yourself in a week 2 lab.
As models become more powerful, it becomes increasingly important to ensure that they perform well in deployment and align with human preferences. In week 3, you'll learn an additional fine-tuning technique called "reinforcement learning with human feedback", which can help ensure your model performs well.
An important aspect of all these techniques is evaluation. Next week, you'll explore some metrics and benchmarks that you can use to determine how well a model performs or aligns with your preferences.
Note that this phase of adapting and aligning application development can be highly iterative.
You might first try prompting engineering and evaluating the output,
Then use fine-tuning to improve performance,
Then review and evaluate the hint engineering again to get the desired performance.
Finally, when you have a well-aligned model that meets your performance needs, you can deploy it to your infrastructure and integrate it with your application.
An important step at this stage is to optimize the model for deployment. This ensures that you get the most out of your computing resources and provide the best possible experience for your application's users.
A last but not least step is to consider any other infrastructure your application needs to function properly.
LLMs have fundamental limitations that are difficult to overcome with training alone, such as inventing information when they don't know the answer, or their limited ability to perform complex reasoning and mathematics.
In the final part of the course, you'll learn some powerful techniques you can use to overcome these limitations.
I know there's a lot to think about here, but don't worry about absorbing it all at once right now. You'll see this architecture again and again throughout the course as you explore the details of each phase.
reference
https://www.coursera.org/learn/generative-ai-with-llms/lecture/21Nwn/generative-ai-project-lifecycle