Inspiration
The inspiration behind Singtel MIVA was to create a groundbreaking customer service experience. Recognizing the diverse and evolving needs of our customers, we aimed to develop an assistant that could not only understand and respond to a variety of requests but also personalize interactions using advanced LLM models.
What it does
MIVA revolutionizes customer engagement by providing personalized, multimodal support. It understands text, visual inputs, and audio across multiple languages, offering tailored solutions based on customer data such as dwelling type, usage patterns, and location. MIVA’s LLM plug-ins enhance its capabilities, making it a dynamic and adaptive assistant.
How we built it
Given the time constraints, we leverage off-the-shelf pre-trained models and API as much as possible to deliver the demo prototype. We built MIVA by integrating state-of-the-art Multimodal Large Language models such as LLaVA (Large Language and Visual Assistant), GPT-4 and Whisper model. In the end, we incorporate it into a Telegram bot.
Challenges we ran into
One of the main challenges was unfamiliarity with Databricks platform in the beginning. Thanks to the organising team, we managed to overcome the problems via office hours and slack.
Accomplishments that we're proud of
We are proud of MIVA’s ability to offer a truly personalized customer experience. Its success in understanding and responding in multiple modalities marks a significant milestone. Moreover, the showcase of LLM plug-ins demonstrates the unlimited possibilities with MIVA.
What we learned
The development of MIVA taught us the importance of multidisciplinary collaboration in LLM projects. We also gained insights into the complexities of LLM and the nuances of multimodal. Balancing model response speed with accuracy was another crucial learning.
What's next for Singtel MIVA
Looking ahead, we plan to expand MIVA’s capabilities to include more plug-ins and use cases, further enhancing its accessibility. Long-term, our goal is to make MIVA a leading example of LLM-driven customer service excellence.

Log in or sign up for Devpost to join the conversation.