Table of Contents:
- Introduction
- Project Overview
- Motivation
- Market Research
- User Survey
- Features
- Architecture
- Getting Started
- Usage
- Contributing
- License
Welcome to the TechDesk - LLM Powered Recommendation System repository! This project aims to provide personalized and expert advice to users who are looking to purchase large-ticket items online, particularly focusing on electronics. The system utilizes advanced natural language processing techniques, including instructor XL for document embeddings and Google Flan XL or GPT-3.5 for conversational interactions.
The TechDesk Recommendation System involves two key components:
- Document Embedding: Expert documents are processed using instructor XL to create meaningful embeddings. These embeddings are then stored in the FAISS (Facebook AI Similarity Search) index for efficient retrieval.
- Conversational AI: The system interacts with end users through Google Flan XL or GPT-3.5, leveraging the stored embeddings to provide personalized recommendations and expert advice in real-time.
The motivation behind the TechDesk project stemmed from the widespread hesitation among users to purchase high-value items online. The lack of confidence in making informed decisions hindered online sales growth. By offering personalized and expert advice through an AI-powered recommendation system, we aim to bridge this gap and empower users to make well-informed online purchasing decisions.
Our initial market research indicated that 62% of individuals were not confident in buying large ticket items online. To delve deeper, we conducted a user survey to understand the reasons behind this hesitation:
In our user survey:
- 91.1% of respondents expressed their willingness to make online purchases if provided with personalized and expert advice on demand.
- 70% of respondents who were interested in online buying had never purchased electronics online.
These findings underscored the potential for a recommendation system that could address users' concerns and provide expert guidance during their online shopping journey.
- Document embedding using instructor XL for efficient storage and retrieval of expert advice.
- Conversational AI using Google Flan XL or GPT-3.5 to engage in real-time interactions with users.
- Personalized recommendations based on user preferences and embedded expert knowledge.
- Seamless integration with e-commerce platforms for an enhanced shopping experience.
The TechDesk recommendation system architecture comprises the following components:
-
Document Embedding:
- Instructor XL processes expert documents to create embeddings.
- Embeddings are indexed using FAISS for fast and accurate retrieval.
-
Conversational AI:
- Google Flan XL or GPT-3.5 powers the interactive conversations with users.
- Embeddings are used to tailor responses and recommendations to user queries.
-
User Interface:
- Intuitive user interface to facilitate user interactions.
- Displays personalized recommendations and advice.
To get started with TechDesk, follow these steps:
- Clone the repository:
git clone https://github.com/Priyanshu-U/TechDesk.git - Install the required dependencies by referring to the installation guide.
- Set up your API keys for Google Flan XL or GPT-3.5 as per the instructions provided.
- Run the system and start interacting with the recommendation engine.
- Launch the TechDesk application.
- Enter your preferences and queries.
- Receive personalized recommendations and expert advice.
- Engage in real-time conversations to clarify doubts and gather information.
We welcome contributions to enhance the TechDesk recommendation system. If you'd like to contribute, please follow the guidelines outlined in CONTRIBUTING.md.
This project is licensed under the MIT License.
We believe that with the TechDesk - LLM Powered Recommendation System, we can empower users to confidently make online purchases, backed by expert advice and personalized recommendations. Happy shopping!