CAiFE uses cutting edge AI technology to create chatbots customized to the needs of any business. This solution democratizes conversational AI and unlocks competitive-advantages for any business, particularly small and medium segments.
It can easily handle common service requests allowing businesses to focus on their core strengths.
Chatbot is available 24 hours and enhances user experience through its accurate and naturalness.
The solution scales during peak demand period by allowing better outcomes for customers and business.
The model could be trained and customized to meet various domain needs and application goals.
You can get your own chatbot by filling out the form below. This video provides a step by step guide on how to fill out the form correctly.
Simple 4-Step Registration Process
Overview of the underlying technology and system architecture deployed
Goal-oriented dialogue systems is a category of dialogue systems designed to solve one or multiple specific goals or tasks (e.g. flight reservation, hotel reservation, food ordering, appointment scheduling). Traditionally, goal-oriented dialogue systems are set up as a pipeline with four main modules: 1-Natural Language Understanding (NLU), 2-Dialogue State Tracking (DST), 3-Dialog Policy Manager, and 4-Response Generator.
NLU extracts the semantic information from each dialogue turn which includes e.g. user intents and slot values mentioned by the user or system. DST takes the extracted entities to build the state of the user goal by aggregating and tracking the information across all turns of the dialogue. Dialog Policy Manager is responsible for deciding the next action of the system based on the current state. Finally, Response Generator converts the system action into human natural text understandable by the user.
Meta learning is learning how to learn. In may cases, we have small data for new and emerging domains. Due to this, high accuracy is not possible. Using data from high data domains, we can train the model to find a good initialization for low data fine-tuning. Starting from a good initialization point allows low data model to reach a optimal point quickly and achieve accuracy levels similar to high data domains.
First the user input needs to be tokenized to find intents and slots. For this, we use a JointBERT model which does two classifications - one for intent and other for slots. Intent classification is done using CLS token based on the entire sentence. For each token output, a classifier is run to classify token into one of the slots.
The state tracker collects all slots generated in a conversation. It has a defined sequence of slots for an intent. Based on the available slots, predefined sequence and business database, it generates a response asking for the next slot in sequence. This response goes back to the user.
There's an offline portion and an online portion. Offline, meta learned models are created for different domains. The model is first trained on high data domains where data is available from Multiwoz. This model is then finetuned on low domain data. In our case, this is synthetically created data and it results in a model that is specific to a particular domain. Many such models could be created, one each for a domain.
When user comes to Caife website, they specify a domain which in turn selects a particular model. This model can be combined with State tracker and response generator to create a business specific URL. The URL is sent to customer via email. The URL can be incorporated in customer's website as a chatbot. The chatbot will run on Caife servers and each business will have a different URL.
Hello and thank you for visiting CAiFE. Please let us know how we could of help.
Testimonies from subject matter experts in varioud domains.
When the pandemic hit us, the human-centric hospital workflow in Chennai, India, was put to test. It also showcased in bad light the efforts we have invested in using AI in patient-sensitive data field. The CAiFE demo is a good proof-of-concept the tech industry should pivot and expand on without relying on data from hospitals.
- Dr. Seshadri Arumugam, M.B.B.S, M.D
Malar Hospital
Many operation activities could be automated in the resource-restricted medical enviroment. CAiFE AI chatbot increases the operational efficiency and engages patients without having to wait for a long queue. The response is accurate, it is a plausible option for Conversational AI in Healthcare Services.
- Dr. Jasmine Tan, M.B.B.S
Perth Royal Hospital
Strong data science experience in various specialization
