Which of these 2 approaches is the best route to learn to build a question answer chatbot?

Question

Quick background on what I am trying to accomplish:

I have been working on a project in my company that requires about 300 people across the world to follow quite a large set of rules and guidelines. instead of sharing documents that people can reference for these rules, I am looking to build a chatbot. This chatbot will take all my documents as inputs and then can be used by the end users to answer any of their questions related to processes and guidelines

What research I have done so far:

Broadly, I believe my 2 main routes are

1. build a chatbot from scratch. I found a good Udemy course on seq2seq architecture that can allow me to build this chatbot
2. Build an application(chatbot) on top of existing algorithms like GPT-3 or BERT

What help I need:

1. I am confused which of the above 2 is the best approach. I guess I need more information on the pros and cons of both.
2. Is 1) is the right approach, is seq2seq too outdated to learn about it now? Should I find a course that uses another architecture?
3. If 2) is the right approach, I am struggling to find a resource that can teach me how to build this chatbot including a simple GUI

Note: I have experience working with CNN and a little bit with RNN as well. I have extensive experience with Python, none with HTML or Java

Answer 1

I would choose option 2 (Build an application(chatbot) on top of existing algorithms like GPT-3 or BERT) for your particular use case, specifically GPT.

For your case, it seems that you are trying to build a chatbot, and for such purposes, it is better to use a large generative language model such as GPT as there are already many resources. GPT, after training has the ability to generate human-like responses to text input which is from what I can tell your use case.

Additionally this won't require much machine learning knowledge as most of the heavy lifting is done by OpenAI for GPT-3 or you can use Transformers from HuggingFace for GPT-2, GPT-NeoX-20B and etc.

For resources:

• How To Build a GPT-3 Chatbot with Python -> you can change the GUI as it utilizes an API implementation
• The AI Chatbot Handbook – How to Build an AI Chatbot with Redis, Python, and GPT -> they use GPT-J-6B but you can change that and follow this as a guide.