Inspiration
As people age, everyday tasks such as remembering medications, identifying scam messages, and staying connected with family become increasingly challenging. Existing solutions are often fragmented across multiple apps and can be difficult for seniors to use. We wanted to create a single, voice-first AI companion that feels natural, supportive, and reassuring while allowing older adults to maintain their independence and giving families greater peace of mind.
What it does
Kindred AI is a voice-first care companion that uses multiple AI agents behind a single conversational interface. It helps seniors manage medications, track missed doses and refills, detect suspicious messages, communicate with family, manage household reminders, answer factual questions from the web, and provide a simple, senior-friendly experience through voice and an intuitive Care Hub.
How we built it
Kindered AI has been built with the help of Codex and gpt-5.6-luna, gpt-realtime-1.5. I was not building alone , I had Codex to help me. Project planning , requirements , features were framed out first. Then, codex created the boiler plate for the backend MCPs , important architectural decisions were taken related to number of agents and voice assistant , usability. 3 dedicated agents wired in an agentic workflow were creating , a router was created to think , decide and take actions by calling the agent. The front end was created in react. For observability Langfuse was used. Tavily API was used for web search.
Challenges we ran into
The app idea was there but the architecture and the usability were the first challenges. Agents communication and real world usability were not clearly understood by Codex so the code created kept deviating from the objective. As more features were added the complexity increased. Persistent memory and state management became a cause of concern which was later on handled. Multiple MCPs and how to deal with the data was also challenging . Finally the demo under 3 mins was a challenge.
Accomplishments that we're proud of
Built a working multi-agent AI system with a seamless voice-first experience. Created a senior-friendly interface focused on accessibility and ease of use. Successfully integrated medication tracking, scam detection, reminders, family communication, and research into one application. Added end-to-end observability using Langfuse for every AI interaction. Demonstrated how agentic AI can solve meaningful real-world problems beyond simple chatbots.
What we learned
Realizing a real world scenario requires a deeper insight into functionality , usability not just technical prowess. Coding agents are good but decision making can be better in terms of architectural choices. Testing and Observability as core principles.
What's next for Kindred AI
I would like to continue working on it and make it more usable. We have got assitants like Alexa or Siri but what if u have an on device assistant which does not share your personal records with the world so privacy is crucial . Kindred AI will attempt to fix the problem which my parents living 7000 km from me are facing everyday.
Log in or sign up for Devpost to join the conversation.