The challenge

The primary challenge for Meta was addressing the widespread dissatisfaction with chatbots. Historically, these bots have been a source of frustration rather than assistance, failing to make meaningful connections with users. Their limitations in scripted conversations, lack of customer-centric planning, and immature technology often resulted in a disengaged and dissatisfied customer base. In a market where conversational AI is expected to drive $112 billion in retail revenue by 2023, the need for a more human-like, engaging chatbot experience was paramount.

The solution

To tackle this, Rehab created Purusha, a cutting-edge AI solution designed to redefine the chatbot experience. Purusha was developed to make bots more human-like, capable of answering complex queries and ensuring smooth interactions across various platforms, including voice, messaging, or augmented reality interfaces.

Key features of Purusha include:

Tone of Voice: Utilizing OpenAI to apply predetermined brand tone-of-voice (TOV) traits to the chatbot's responses. Users could adjust these traits to see their live impact, ensuring the chatbot's personality aligned with the brand's essence.

Intent Matching: This feature involved scoring user utterances against intended responses, reducing human input errors, and providing AI-generated alternative responses for better accuracy in understanding user needs.

Conversation Fallback: Employing prompt engineering for fallback conversations outside of the standard chatbot flows, enabling more natural and less robotic interactions.

The innovation

Purusha was not just a chatbot; it was a personalized conversation builder. It enabled open-ended interactions with GPT-3, Blender Bot, and Sentence Transformer, generating dialogues akin to a movie script. This approach allowed the chatbot to adapt to a variety of scenarios, ensuring a dynamic and engaging user experience.

The conversation model was designed to first check user messages against a list of 'intents', allowing the brand to intercept and act on specific user desires effectively.

For instance, in the case of a grocery brand, Purusha could be configured as an expert cook, engaging users in discussions about different cuisines and cooking tips. Specific intents like 'Carbonara' could be trained into the system, allowing the chatbot to surface custom messages written by copywriters for a more authentic interaction. Any questions not matching predefined intents would be seamlessly handled by GPT-3, generating human-like responses.

Crucially, Purusha's architecture was designed with flexibility in mind, accommodating the unique needs of different brands. This was achieved through 'Infrastructure as Code', which allowed for the creation of a new Sentence Transformer model and API for each client, facilitating the deployment of unique chatbots for different brands on distinct API links.

The results

Meta's collaboration with Rehab and the implementation of Purusha marked a significant transformation in the realm of chatbot interactions.

By humanizing these AI interactions, Meta not only enhanced user engagement but also set a new standard for customer-brand interactions in the digital domain.

Purusha's innovative features and adaptable architecture heralded a new era in conversational AI, where chatbots are not just tools for answering queries but are companions that develop and nurture customer relationships, today in messaging platforms and potentially tomorrow in the metaverse.

‍