Inspiration
Better healthcare, distant families, and falling birth rates mean we are staring at a world where a big chunk of the population is going to be seniors(60+). Among these seniors, loneliness is a big problem. So this started our thought process on how Generative AI can help us better connect with our Elderly.
What it does
Saathi is an app that interacts and is a personal assistant meant to be always there! even when some human is not there. While we strongly believe that the "Human Touch" is important, sometimes it is not possible, so Saathi is there for you. The app is built from the ground up with the elders in mind. It answers any questions they may have and reminds them of things they may have forgotten. It also identifies people around them and provides a video memory to help them recollect who they may be interacting with. It is highly personalized and learns about the user with every interaction.
How we built it
We used chromdb as vector-store to store healthcare data, personal information and used langchain to build a RAG model on top of it. We used the GPT-4o for interaction, Eleven labs for personalized audio, and Groq for fast speech-to-text inference. Lastly, we integrated the twelve labs' multimodal API for video querying via text and image.
Challenges we ran into
Integrating multiple APIs and DBs proved to be a challenge. Much trial and error was needed to perfect the LLM prompts to prevent dangerous misinformation and make our AI responsible. Integrating agents was another challenge, and we are still working on this.
What's next for Saathi.AI
Doctor integration to provide further context on medication reminders. Agents for automating everything. Rich video integration using Twelve Labs
Log in or sign up for Devpost to join the conversation.