AI Sales Agent

Inspiration

The inspiration for the AI Sales Agent project came from the need to streamline customer interactions in e-commerce. Human-in-the-loop based sales conversations consume a lot of time and resources. We aimed to develop a solution that could handle product inquiries, assist in decision-making, and take orders efficiently, all while offering a conversational experience to users.

What it does

The AI Sales Agent is a conversational system that:

1. Pitches products to users based on their preferences, adding follow-up questions wherever needed.
2. Retrieves product information (e.g., categories, descriptions, prices) from a vector database.
3. Takes customer orders by creating an entry with their name, assigning a unique user ID, and storing the selected product data in a database.
4. Utilizes Groq's llama-3.3-70b-versatile for conversational AI and Jina embeddings for product matching.

How we built it

We built the AI Sales Agent using the following components:

• Groq API: Used for implementing conversational AI with the llama-3.3-70b-versatile model.
• Jina Embedding API: Employed for generating embeddings for product data.
• FAISS Vector Database: Enabled efficient retrieval of product information.
• SQLite Database: Designed to store customer orders and track unique user IDs.

Challenges we ran into

1. Ensuring seamless integration between multiple APIs and databases.
2. Optimizing the performance of the conversational AI model to handle diverse queries.
3. Designing a scalable database structure to manage user and product information.
4. Addressing edge cases in conversational flow to enhance user experience.
5. Meticulously crafting a system prompt for the LLM to ensure conversational logic.

Accomplishments that we're proud of

• Successfully integrating Groq's high-performance AI model for natural conversations.
• Implementing an efficient and scalable retrieval system using FAISS.
• Creating a user-friendly interface for handling product inquiries and orders.
• Streamlining the process of embedding generation and matching using Jina API.
• Achieving a latency of <2 seconds for response generation and text-to-speech conversion.

What we learned

• How to effectively use large language models like llama-3.3-70b-versatile for real-world applications.
• The importance of optimizing retrieval systems for quick and accurate responses.
• Techniques for building agentic RAG based tools using frameworks like Langchain.
• Strategies for managing and scaling databases in AI-driven applications.

What's next for AI Sales Agent

• Enhancing the conversational flow by incorporating sentiment analysis.
• Integrating a payment gateway for seamless transactions within the agent.
• Expanding the product database to cover more categories and use cases.
• Introducing multilingual support to cater to a broader audience.

Built With

• faiss
• groq
• jina
• langchain
• llama
• python
• smallest.ai
• sqlite
• streamlit