Being two university students, we have often found the need to become a lender or lendee. This ranges from buying a concert ticket for a friend because he is not able to do it at that moment to being on the receiving end to buy a new textbook. Lending money has always been a very uncomfortable topic for many until now. Our goal was to fix the issue which is intrinsic to micro-loans, the fact that it is always a gamble whether or not you will receive your money back. In addition, we see the potential for P2P loans to help break the poverty cycle and generate an economic multiplier effect for credit-constrained households.
What it does
Using Open Bank APIs, our platform allows for FIAT-to-FIAT lending, while maintaining a record of all transactions referring to the loan on the blockchain. As such, we can reach a wider amount of user globally who may not be able to take part in crypto P2P lending initiatives. Also, by keeping an account of the loan on the blockchain, we leverage the immutability aspect of the technology to provide a way for users to hold each other accountable on the loans. In addition, we have created a decentralized credit score, which allows for people who normally would not have access to a loan to be able to make use of credit.
A key aspect is the machine learning technology utilized in the loan confidence meter. We trained a Support Vector Machine (SVM) to predict if an individual will repay a certain loan (as a function of time, credit score, amount and interest rate) as a binary classification problem. The algorithm then utilizes Platt scaling to convert this binary classification into a probability distribution as a function of distance from the hyper-plane. The result is a level of "confidence" that the user will pay back the loan. A further feature of the machine learning side of the project is that we generate a suggested interest rate as a function of time, credit score and amount. The algorithm works in a "online" environment, meaning that it has the ability to retrain as users keep using the platform. This means that machine predictions will keep on improving with time as more users utilize the service, allowing for the natural improvement of our probabilistic models. With more data and a wider feature space the technology could very feasibly evolve to utilize a deep learning neural network to appreciate the intricacies which are intrinsic in human behavior psychology and manifold learning.
How we built it
The backend is written in node.js and utilizes express as a web engine and jade as a templating engine. Blockchain stuff here
Accomplishments that we're proud of
We are proud that the two of us managed to participate to the event regardless of our other commitments and manage to complete a project which we believe to reflect our best abilities in such a short time frame. We are proud that we managed to find the time to convey our passion for development and improvement to a large group of talented and driven individuals. We are proud that we managed to solve great technical challenges involved in integrating bank systems, blockchain and machine learning to the best of our abilities.
Yakko Majuri, King's College London BSc Business Management (yr 1) Blockchain, Frontend, Backend (MongoDB & Express) Yakko is an Ethereum developer and co-founder of he education platform BlockchainBH in Brazil. With IGTI, he worked as a Visiting Scholar in the development of the first MBA with a focus on blockchain in Brazil, and currently works as Project Lead for London Blockchain Labs. Despite not coming from a technical background, Yakko spends the majority of his days programming, while also being a Business Management Student a KCL.
Federico Barbero, King's College London MSci Computer Science (yr 1) Machine Learning, API, Backend (API & Express) Federico is a researcher in cybersecurity and machine learning at King's College London. His research interest involves conformal prediction theory applied to ML and cyber security, metric spaces and topology applied to quarantine techniques and information geometry. He works part-time as a Python/VBA/Tableau teacher/developer at Finisterre Capital.