Build Your LLM-Powered Personal Website

Introduction

Since ChatGPT shocked the world with its capability, AI started to be utilized in numerous fields: customer service assistant, marketing content creation, code assistant, travel itinerary planning, investment analysis, and you name it.

However, have you ever wondered about utilizing AI, or more specifically Large Language Model (LLM) like ChatGPT, to be your own personal AI assistant on your website?

A personal website, usually also called a personal web portfolio, consists of all the things we want to showcase to the world, starting from our short biography, work experiences, projects we have done, achievements, paper publications, and any other things that are related to our professional work. We put this website live on the internet and people can come and see all of its content by scrolling and surfing the pages.

What if we can change the User Experience a bit from scrolling/surfing to giving a query? What if we add a small search bar or widget where they can directly ask anything they want to know about us and they’ll get the answer immediately?

Let’s imagine a head-hunter or hiring manager who opened your personal website. In their mind, they already have specific criteria for the potential candidates they want to hire. If we put a search bar or any type of widget on our website, they can directly ask what they want to know about us. Hence, improving our chances of being approached by them. This will also let them know that we’re adapting to the latest technology available in the market and surely will increase our positive points in their scoring board.

In this article, I’ll guide you to build your own LLM-Powered Personal Website. We’ll start by discussing several FREE-ly available LLMs that we can utilize. Then, we’ll go into the step-by-step process of how to build our personal AI assistant by exploiting the LLM capability as a Question and Answering (QnA) module. As a hint, we’ll use one of the available task-specific models provided by AI21Labs as our LLM. They provide a 3-month free trial worth $90 or 18,000 free calls for the QnA model. Finally, we’ll see how we can put our personal AI assistant on our website.

Without wasting any more time, let’s take a deep breath, make yourselves comfortable, and be ready to learn how to build your own LLM-powered personal website!

Freely Available LLMs

The main engine of our personal AI assistant is an LLM. The question is what LLM should we use?

There are many variants of LLM available in the market right now, starting from open-source to closed-source LLM. There are two main differences between open-source and closed-source. Open-source LLMs are absolutely free but you need to host it by yourself. On the other hand, closed-source LLMs are not free but we don’t need to host it by ourselves, we just need to call an API request to utilize it.

As for open-source LLM, the go-to LLM for a lot of use cases is LLaMA-2 by Meta AI. Since LLM consumes a large amount of GPU memory, in practice we usually perform 4-bit quantization to reduce the memory usage. Thanks to the open-source community, you can now directly use the quantized version of LLaMA-2 in the HuggingFace library released by TheBloke. To host the LLM, we can also utilize a very powerful inference server called Text Generation Inference (TGI).

The next question is whether there are freely available GPU machines out there that we can use to host the LLM. We can’t use Google Colab since we want to host it on a server where the personal website can send API requests to that server. Luckily, there are 2 available free options for us: Google Cloud Platform and SaturnCloud. Both of them offer free trial accounts for us to rent the GPU machines.

Open-source LLM like LLaMA-2 is free but it comes with an additional hassle which is to host it by ourselves. In this article, we’ll use closed-source LLM as our personal AI assistant instead. However, most closed-source LLMs that can be accessed via API are not free: GPT-3.5 and GPT-4 by OpenAI, Claude by Anthropic, Jurassic by AI21Labs, etc.

Luckily, AI21Labs offers $90 worth of free trial for us! Moreover, they also provide task-specific models that are charged based on the number of API calls, not based on the number of tokens like in other most closed-source LLMs. This is surely very suitable for our use case since we’ll have long input tokens!

Let’s dive deeper into AI21Labs LLM, specifically the QnA model which we’ll be using as our personal AI assistant!

AI21Labs QnA LLM

AI21Labs provides numerous task-specific models, which offer out-of-the-box reading and writing capabilities. The LLM we’ll be using is fine-tuned specifically for the QnA task, or they called it the “Contextual Answers” model. We just need to provide the context and query, then it will return the answer solely based on the information available in the context. This model is priced at $0.005 / API request, which means with our $90 free trial account, we can send 18,000 API calls! Isn’t it amazing?

 Without further ado, let’s start building our personal AI assistant!

1.    Create AI21Labs Free Trial Account

To use the QnA model, we just need to create a free trial account on the AI21Labs website. You can follow the steps from the website, it’s super easy just like creating a new account on most websites.

2.    Enter the Playground

Once you have the free trial account, you can go to the AI21Studio page and select “Contextual Answers” under the “Task-Specific Models” button in the left bar. Then, we can go to the Playground to test the model.

Inside the Playground of the QnA model, there will be 2 input fields and 1 output field. As for input, we need to pass the context (the knowledge list) and the query. As for output, we’ll get the answer from the given query based on the context provided. What if the answer doesn’t exist in the context? This model will return “Answer not in documents.” as the fallback.

3.    Create the Knowledge List

The next and main task that we need to do is to create the knowledge list as part of the context input. Just think of this knowledge list as the Knowledge Base (KB) for the model. So, the model is able to answer the model only based on the information available in this KB.

4.    Test with Several Queries

Most likely, our first set of knowledge is not exhaustive. Thus, we need to do several iterations of testing to keep expanding the list while also maintaining the quality of the returned answer. We can start by creating a list of possible queries that can be asked by our web visitors. Then, we can add several answers for each of the queries inside the knowledge list.

 Pro tip: Once our assistant is deployed on our website, we can also add a logger to store all queries and responses that we get. Using that log data, we can further expand our knowledge list, hence making our AI assistant “smarter”.

5.    Embed the AI Assistant on Our Website

Until now, we just played with the LLM in the Playground. However, our goal is to put it inside our web portfolio. Thanks to AI21Labs, we can do it easily just by adding the JavaScript code inside our website. We can just click the three-dots button in the top right of the “context input” and choose the “Code” option. Then, a pop-up page will be shown, and you can directly copy and paste the JavaScript code into your personal website. That’s it!

Conclusion

Congratulations on keeping up to this point! Hopefully, I can see many new LLM-powered portfolios developed after this article is published.

 Throughout this article, you have learned how to build your own LLM-powered personal website starting from the motivation, freely available LLMs with their pros and cons, AI21Labs task-specific models, creating your own knowledge list along with some tips, and finally how to embed your AI assistant in your personal website. See you in the next article!

Author Bio

Louis Owen is a data scientist/AI engineer from Indonesia who is always hungry for new knowledge. Throughout his career journey, he has worked in various fields of industry, including NGOs, e-commerce, conversational AI, OTA, Smart City, and FinTech. Outside of work, he loves to spend his time helping data science enthusiasts to become data scientists, either through his articles or through mentoring sessions. He also loves to spend his spare time doing his hobbies: watching movies and conducting side projects.

Currently, Louis is an NLP Research Engineer at Yellow.ai, the world’s leading CX automation platform. Check out Louis’ website to learn more about him! Lastly, if you have any queries or any topics to be discussed, please reach out to Louis via LinkedIn.