Inspiration
In today's fast-paced world, many seniors experience deep loneliness and isolation, despite the abundance of digital communication tools. The lack of genuine human interaction leaves their rich life stories and experiences fading into silence, impacting their emotional well-being. Our vision is to offer relief by providing a constant, caring presence that listens, shares in their joys and sorrows, and offers comfort whenever needed. This inspired the creation of NATSU AI, bridging the gap between technology and heartfelt connection to ensure every senior feels valued and supported.
What it does
NATSU AI serves as a human-like AI companion for senior citizens, offering several features:
Health Monitoring: reminds users to take medications, stay hydrated, and maintain healthy routines through meal checks and a physical exercise tracker.
Daily News Updates: Fetches top 5 news articles to the user, allows them to select those articles and asks follow-up questions for deeper understanding.
Conversational Memory with Visual Recall: Users can share memories, thoughts, and photos and ask about those moments to get a friendly response that recalls the story and shows the original image.
Emotional Support: Provides empathetic check-ins to combat loneliness.
Engaging Storytelling: Shares stories and engages in thoughtful discussions.
Dates & Events Reminder: Notifies users of important events like birthdays and anniversaries.
How we built it:
To build NATSU AI, we combined several state-of-the-art technologies in a thoughtful and integrated architecture:
We used Gemini 1.5 Flash for feature classification and advanced NLP tasks, enabling the system to understand and categorize user inputs effectively.
Our team developed a custom psychometric algorithm that evaluates emotional metrics such as arousal, trust, and valence. This algorithm was based on extensive research and plays a key role in generating emotionally aware responses.
For memory and personalization, we implemented Graph RAG using LangChain, creating knowledge graphs from user interactions. This enables NATSU AI to remember events, preferences, and memories.
These graphs are then used by a second stage RAG pipeline which connects to Gemini 2.0 Flash, responsible for generating the final human-like and emotionally adaptive responses.
For speech-to-text (STT) processing, we integrated Deepgram, providing accurate and fast transcription.
For text-to-speech (TTS), we are using ElevenLabs, ensuring the voice of NATSU AI sounds warm, clear, and natural.
This combination of emotional intelligence, memory-driven personalization, and real-time speech interaction makes NATSU AI a truly human-like and compassionate companion.
Challenges we ran into:
Implementing Graph Retrieval-Augmented Generation (Graph RAG) posed significant challenges due to its novelty and complexity, particularly when using Gemini embeddings (model - models/embedding-001). Fine-tuning the proper RAG extraction and determining optimal thresholds required extensive testing cycles. We frequently encountered issues with the system either missing critical information from RAG or overly relying on it, even for minor queries. This optimization was highly time-consuming. Additionally, developing our psychometric algorithm to accurately determine emotional metrics such as arousal, trust, and valence was a substantial research effort. Achieving this novel approach required dedicated exploration and refinement.
Accomplishments that we're proud of
We have got traction and validation through the Sac State Startup Challenge where we won 2nd, validating both product and market fit. Through this traction, we conducted 100+ user interviews, secured 2 pre-sale customers at $50 each, and built a 50+ member waitlist showing strong and growing interest.
We launched our website at natsu.ai(show-case only) and partnered with Purple Door Finders, unlocking access to 60,000+ active seniors actively seeking wellness and companionship solutions.
Through the last 48 hours we built a fully functional MVP, featuring voice input, vision processing, memory recall, local data storage, and multi-threaded backend for low-latency response execution.
Developed an emotionally adaptive AI companion that feels genuinely empathetic and supports natural conversations, including news reading with follow-up interaction to boost user engagement and mental stimulation.
What we learned
This project taught us the importance of designing with empathy, especially when building technology for seniors who may feel isolated or overwhelmed by digital tools. We learned how vital it is to strike a balance between simplicity and functionality, ensuring every feature feels intuitive and purposeful. Collaborating across disciplines—from healthcare to user-centered design—enabled us to create a solution that’s not just intelligent but also genuinely caring and approachable. Most importantly, we embraced challenges as part of the journey—iterating quickly, listening closely to feedback, and continuously adapting to meet real human needs.
What's next for NATSU AI
10k in pre-seed, develop and ship product by end of May and get 10,000 paying customers by end of the year and 1m in total funding for a strategic plan.
Log in or sign up for Devpost to join the conversation.