Training a Custom GPT Model for Call Transcripts Analysis

GPT builders
1

Hi OpenAI community,

I hope this message finds you well. I’m reaching out to seek your expertise and guidance on a project I’m currently working on at my company. We have a vast repository of call transcripts between users and sales agents, and I’m eager to leverage the capabilities of ChatGPT through the API to analyze and understand the key aspects of these conversations.

My goal is to create a custom GPT model that can assist in extracting valuable insights from these transcripts, such as identifying the most common user problems, understanding sales agents’ responses to specific queries, pinpointing frequently asked questions, and discerning overarching themes in user inquiries.

Here are some specific areas where I’d appreciate your insights:

1. Training Process: What would be the optimal approach to train ChatGPT on our call transcripts using the API? Are there specific considerations or best practices that we should be aware of when tailoring the model to our specific use case?
2. Question Formulation: How can I structure queries to the model effectively to extract information about the biggest problems faced by users, sales agents’ responses to specific questions, frequently asked questions, and overarching themes in user queries? Are there certain techniques or strategies that have proven successful in similar applications?
3. Fine-Tuning: Are there recommended methods for fine-tuning the model to improve its performance on our specific dataset? What parameters or hyperparameters should I focus on to enhance the model’s understanding of our call transcripts?
4. Bot Integration: Once the model is trained, how can I integrate it into a bot that can assist in the analysis of new transcripts and provide valuable insights in real-time?

I truly appreciate any insights, advice, or resources you can share to help me navigate this project successfully.

1 Like
2

My first thought is to ensure you have a pre-existing written agreement with all of your users and sales agents to collect their call transcripts over the period they were collected and to ensure that agreement allows you to submit those details to a third party (in this case case OpenAi… and potentially it could be folded into the overall ChatGPT knowledge and impact responses to other ChatGPT users). For example, my company does not permit any data to be used in this manner.

This should be confirmed and formally approved by your company before you start on your plans.

A second key consideration is: What checks will be done to ensure no personally identifying or sensitive data will be used as input. You mention having a “vast repository” which increases the risk of inappropriate data being used as input.

3

on youtube: " Step-by-Step Guide: Build Your Own Custom GPT for Qualitative Data Analysis"

watched this the other day, was pretty helpful for a beginner like me

4

Using an example transcription and inputing that into our custom built tc copymaster chat A.I, it was able to summarize the transcript in an organized way that would allow anyone who maybe missed or didn’t catch things in the interview to fully understand what the interview was about, who was in it and key points and goals.

5

Although it was able to accurately summarize the transcript, it left some key points unfinished which can be confusing for some.