Inspiration
i was inspired by how the Qwen3 model was trained and it could be done and by how many businesses in Africa want to intergrate AI but they are worried about privay and data security.
What it does
introducing Serget AI power network. this si a decentralized peer to peer blockchain network that is based on th etherium network and provides a connection to different nodes and all the nodes on the network contribute comnpute power towards a given task
How we built it
we used a private etherium network called geth as our blockchain network this decesion was done to redefine the gas fees. when the owner uploads a model to the network it is sharded and chunked into small parts which are split across nodes in the network it being blockchain it encourages redunduncy to ensure full uptime when an inference qurey is made we load the model and serve it with vllm since it has native muti node support types of nodes bootstrap worker mobile
the current setup also does decentralized file storage for companies with data storage needs at a cheaper price
Challenges we ran into
model sharding and chunking smart contract developmant hosting the project and honestly the project isnt working well but we belive this si the future for AI and specifically AI in Africa
Accomplishments that we're proud of
wifi -- bluetooth hybrid support for semi internet access reliance provision for mcp intergrations with tools and datasources distributed inference and being able to provide this service to exisiting services and apps to both the public and businesses we had to build an intelligent algorithm for splitting the model depending on your power
What we learned
ive learnt that blockchain isnt just about crypto I learnt solidity we learnt different frameworks and techniques of building such asn ambitious project ive learnt that most probably implementing such an ambitious idea would need more than a month and a first year sitted in a lab
What's next for Serget AI power Network
predictive node analysis -- being able to detect anomalies and failures in diferent nodes and also predicting storage capacity limit date providing this to both users and businesses being able to train modles from this network since you can provide data and nodes can provide compute power no matter how small to be honest this project has the potential to be great with more hands on since im literrally working alone
Log in or sign up for Devpost to join the conversation.