VAST

Inspiration (Inspiration, How the solution is special / novel / unique)

With the rise of AI chat bots, we felt that we could build on it as a base, and fit it into multiple scenarios. We thought of adding additional voice recognition and text-to-speech functionalities, and combining them with a natural language AI model, to provide users with a more human-like experience. As the current market does not contain many offline tools, we thought of fitting our solution into this hole, by allowing full offline capabilities with a built-in database.

What it does (Problem that you are trying to solve and its significance,

How the product is meant to be used) VAST is a offline voice assistive search tool that is designed to allow people, especially those with disabilities, to search the internet without having to type anything. For example, blind people are now able to search the internet without having to see what they are typing with the use of speech-to-text and text-to-speech. In addition, in areas without internet connection, people are able to search for essential information by asking VAST, which will then use its own data storage to generate a reply for the user.

How we built it (The technologies or techniques used)

For the text-to-speech and speech-to-text part, we used python. Next, for the brain, we used a natural-language-processing model known as llama2 to generate and summarise responses to the user's queries. In addition, we used python and google for online enrichment, while also using fastapi and flask for inter-connectivity components.

Challenges we ran into (Key challenges)

Porting the whole system offline, catering to 3rd world countries with poor network connectivity or expensive data costs, posed a significant challenge for us. Both the data set and the voice-to-text functionality had many difficulties sourcing accurate information and processing data, without relying on reliable connectivity to cloud databases that are readily available in 1st world countries like Singapore. Additionally, current offline libraries are not as accurate or reliable as online solutions, so we had to slowly tweak it through trial and error, to make it as accurate as possible.

What's next for VAST (The product as a business and/or at scale)

We hope that this product will be made portable enough to be installed on mobile phones, for people from less developed countries to use without data. As this product does not require server hosting, non-profit organisations can distribute it easily without additional costs

Built With

• c
• fastapi
• google
• llama
• natural-language-processing
• python