How I made my own chatbot using RAG

In this post, I'm going to talk about how I made my own chatbot for my personal web portfolio here..

When I first stumbled across the concept of RAG, I wondered how this is any different than just training ChatGPT to give answers based on data given in the prompt.

Here's why RAG is important:

1. More developer control: RAG gives the developer more control over information sources and how it is presented to the user. They can restrict sensitive information and also provide the latest information to users.
2. Cost Effective: RAG is much cheaper than training a model to operate in a domain-specific area.
3. Use-case specific: RAG provides output based on only the context provided to it. This enables developers to create tailor-made models to only respond to domain-specific questions and not give vague responses outside the model's area of expertise.

Now let's get to the fun part - actually making a chatbot!

I started out by creating the context for my chatbot. I asked chatGPT to write a 1000-word text to train a RAG model based on my resume.

Once, I had the context, I used the chatOpenAI API from the langchain/openai library to define my model. I decided to go with the gpt-3.5-model.

I created a prompt asking the LLM to answer questions as if it were an AI version of me, using the data given in the context. I played around with the temperature and prompt for a bit until I finally got satisfactory results.

This was a fun project that taught me about RAG architectures and gave me hands-on exposure to the langchain library too.

Make sure to check out my website and try the chatbot for yourself here!