Skip to content

ajruttam/Udacity-GenAI-Phase2-Project2

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

4 Commits
 
 
 
 
 
 

Repository files navigation

Udacity-GenAI-Phase2-Project2

Build Your Own Custom Chatbot

The main objective of this project is to build a customized chatbot capable of answering questions using a customized database. That is, a data set that was not used during model training.

Rubrics

1. Choose a Dataset and Explain the Scenario

In this project, I choose 2023_fashion_trends.csv as my dataset provided in the data folder of this module. I am using it because the model I will be using is gpt-3.5-turbo-instruct which was trained till 2021 and it might not know about the 2023 fashion trends.

To improve the question answering capability of our tool, we employ the Retrieval Augmented Generation technique. This technique complements the prompt with contextual information from the dataset, allowing the model to provide more accurate and relevant answers to the questions posed to it.

2. Prepare the Dataset for the Custom Query Process

The chosen dataset was loaded into a pandas dataframe with a column called "text". This column should contain all of your text data, separated into at least 20 lines and then transformed into a DataFrame so that embeddings could be created in the next step. The objective of this was to demonstrate the correct functioning of the project, since OpenAI's GPT-3.5-instruct has information until 2021.

3. Perform the Custom Query Process

To perform the customized query, embeddings were created using the OpenAI API and saved in the csv file. With this, a customized search was carried out with an OpenAI Completion Model.

4. Write Questions to Demonstrate Custom Performance

To evaluate the results of the developed project, a series of questions were asked, and each was answered both without context and with context.

About

Build Your Own Custom Chatbot

Resources

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published