Inspiration
Home and auto maintenance has been inaccessible to most people who own them for very long. Fixing household problems and car issues has often a high barrier of knowledge, skill and not to mention, can turn dangerous to users' life and health with minor oversight. Overall, while there exists a lot of data on specific parts and instructions needed to fix these issues, the accessibility to these resources remain the same antiquated way of browsing through manuals and utilizing guesswork to solve issues bearing a significant risk on health. There is no app that is currently available to the masses at their convenience to use to solve daily problems around their house or their vehicles.
Triton Learning is an LLM (Large Language Model) fine-tuned to a large quality dataset of queries and solutions to relevant problems, that solves users' specific questions about their car and home issues the most intuitive way possible - through conversation. Fine-Tuning allows the LLM to have answers of the specificity that users need to ensure they are able to get a starting point on fixing things in a secure, easy manner.
What it does
Fine tuned LLM that answers problems trained on a dataset of similar, closely associated problems in a chatbot, virtual assistant user experience.
How we built it
Front-end: React Back-end: MongoDB Atlas, GCP, Express.js, Node.js
Challenges we ran into
Fine tuning LLM through Google Cloud.
Accomplishments that we're proud of
Connecting the back-end to the GCP Vertix AI to generate answers. Creating API endpoints to access the AI's functionalities.
What we learned
Have fun
What's next for Triton Learning
Further experiment with Fine-Tuning to provide more accurate answers and research on database quality.
Log in or sign up for Devpost to join the conversation.