Assistant Extensions
We enhance our Assistants to perform like a trained expert in your domain
Function Calling
What is Function Calling?
Function enhances generative language models by enabling it to trigger external functionality in response to an input message. This enables the model to perform additional operations, such as API requests, database queries, or executing any kind of pre-existing application. Our Conversational Assistants come natively equipped with function calling for product retrieval and shopping cart interaction.
How does Function Calling help?
Function calling lets the Assistant perform specific tasks by invoking predefined functions. For example, when a customer asks about their order status, the assistant can call a function to query the order management system and provide a real-time update. This method ensures that the assistant delivers precise and actionable information, improving the efficiency and effectiveness of customer service.
Order Mangement
When a customer inquires about their order status, the assistant can invoke a specific function to retrieve real-time updates, track shipments, process returns, modify orders or any other online service your platform offers.
Sales Consultation Bookings
When a customer requests personalized assistance, the assistant can call functions to look up the appropriate expert Sales Associate, check their availability, schedule appointments, and send confirmation details with a summary of the enquiry to both the customer and the associate.
Account Management
The assistant is extended to handle tasks such as updating personal information, resetting passwords, and managing payment methods. When a customer needs to change their shipping address or review their order history, the assistant can execute the necessary functions to make these updates securely and efficiently.
Retrieval Augmented Generation (RAG)
What is RAG?
RAG systems enhance generative models by integrating information retrieval from external data sources. When responding to an enquiry a stand-alone model relies on its training data to generate its response, which might not be sufficiently up-to-date or detailed enough to be useful. A RAG system actively retrieves relevant information from additional data sources and feeds it into the response. Our e-commerce Assistants come natively equipped with a RAG System for item descriptions, enabling them to answer product questions precisely.
Expert Support
RAG can be used to access handbooks, manuals or documentation for specialized fields. When a customer asks a complex support question, the chat assistant can retrieve and generate accurate answers from these expert resources. This ensures that customers receive precise, authoritative guidance on using or troubleshooting products.
How does RAG help?
A RAG system benefits the Assistant by taking into account external information bodies, such as expert documentation, reviews or other large text corpus’ from external data sources to generate accurate and detailed responses to customer inquiries. Unlike traditional chat assistants that depend only on their pre-trained internal knowledge, this approach ensures the assistant delivers precise, current, and relevant information, significantly enhancing the response value to the customer.
Product Questions
RAG enables the chat assistant to pull up-to-date and detailed product information. When asked about specific features, availability, or comparisons between products, the assistant can generate responses that reflect the latest and most relevant details.
FAQs
RAG can efficiently retrieve answers from a database of frequently asked questions (FAQs). When customers pose common inquiries, the chat assistant can quickly access and provide accurate responses drawn from the most relevant FAQ entries. This not only saves time for both customers and support staff but also ensures consistency and accuracy in the information provided.
Finetuning
What is Finetuning?
Finetuning involves adjusting the Assistant’s model weights by further training it on specific datasets to tailor its performance to a particular task or domain. Unlike a generic model that has broad but general knowledge, fine-tuning customizes the model to understand and generate more accurate and relevant responses for specific applications. This process leverages the model's existing capabilities and refines them to meet specialized needs, improving its effectiveness and accuracy in targeted scenarios.
How is this different from a RAG System?
Fine-tuning involves embedding domain-specific knowledge directly into the model by training it on specialized datasets, which allows for quick, nuanced responses and function calls based on the training material. In contrast, RAG systems dynamically retrieve current information from external sources, ensuring responses are accurate and relevant. While fine-tuning provides a deeper internalized understanding of specific topics, RAG systems ensure that the information remains up-to-date. Together, they complement each other by combining trained expertise with real-time accuracy, resulting in comprehensive and highly reliable responses.
Domain Expertise
To deliver expert-level support for complex queries, fine-tuning involves training the assistant on specialized datasets, such as technical documents and industry best practices. This enables the assistant to provide detailed explanations and precise troubleshooting, catering to users with advanced, domain-specific needs.
How does Finetuning help?
Finetuning a chat assistant on your specific data - such as product documentations, customer interactions, and support tickets - makes the assistant better at understanding and responding to customer queries in the context of the your unique offerings and terminology. As a result, the assistant can provide more accurate product recommendations, call functions and resolve issues more efficiently, and deliver a more personalized and effective customer service experience.
Industry Vernacular
Fine-tuning can be utilised to incorporate industry-specific terminology into the assistant’s training. In a real estate platform, this enables the assistant to understand and use terms like “escrow” and “closing costs,” ensuring smooth and professional interactions with knowledgeable customers.
Product Recommendations