Aion | Devpost

Person in India using our chatbot

Inspiration: Came from wanting to make the most powerful tool (A.I.) accessible to the whole world in the cheapest possible way.

What it does: anyone, including a flip phone user, can text our phone number and receive access to an a.i. agent who will answer any question and even call you to assist you with tutoring and/or medical advising. Our LLM automatically simplifies complex concepts into local dialects, ensuring that life-changing information is accessible even at the lowest literacy levels.

How we built it:

We used multiple methods for communicating with our customers in India and other distant countries. By utilizing an Android phone and a new SIM card, I opened up an SMS gateway that allowed my laptop (and later a Vultr) server to receive and process the messages. The server was built with Claude Opus 4.7. The backend LLM was GPT5.5-mini, and ElevenLabs Multilingual for voice.

Challenges we ran into: navigating the bureaucracies of having an international number, sending out automated messages, and calls. Getting the correct interface to our users and the most accurate knowledge system to train our LLM.

Accomplishments that we're proud of: we were able to use Fast-Fourier Transforms to record the heartbeat of the user through the microphone of a flip-phone and filter out other noise, allowing our AI to know how high the user's pulse is. Making a Nokia 105 2G phone number have access to an LLM with call features also available. Our proudest accomplishment is onboarding an international user from India onto our app.

What we learned

During our hackathon, we learned how to effectively implement each AI voice within a third-party platform. We learned to narrow our product to most effectively solve the problem at hand. We learned how to utilize LLM's for deep research and creating the necessary cultural context for Aion models.