Inspiration
AI agents and autonomous trade are rapidly changing how people and businesses interact with each other financially. However, how do we know we can trust the other party, be it a person, organisation or AI agent?
Whilst current payment platforms focus on just facilitating a transaction, we believe that in the future we want to add an intelligent layer of trust within digital payments.
This led us to create PAYPAL 2.0 - Trust Before Transactions, Intelligent Payments.
What it does
How we built it
An AI powered Finance Trust Intelligence platform which evaluates the trustworthiness of individuals, organizations, service providers, and AI agents.
It makes use of AI agents to perform analysis on Identity Evidence, Transaction Data, Fulfilment Data, Business Data and Potential Risk Indicators.
An explainable intelligence trust result will be provided, along with transaction Risk Assessment and recommended payment protection methods.
The platform includes:
Identity Intelligence to provide analysis on Identity Evidence and identify areas where verification is required.
Transaction Intelligence to provide analysis on Transaction Data, Fulfilment History, Refunds and Anomalies.
Risk Intelligence to identify potential risks and provide explanations.
Trust Intelligence to provide an explainable trust result using the evidence available.
Adversarial Intelligence to question the conclusions provided by AI agents and identify any inconsistencies or unbacked conclusions.
Payment Policy Engine to provide deterministic payment policies.
Integration with PayPal to provide an authorised payment using the sandbox.
Challenges we ran into
One of the biggest Challenges We Faced was designing a trust system that would make sense. We couldn’t just assume that if there wasn’t any data then they were fraudulent.
We had to differentiate between trustworthiness, transaction risk, verification and confidence in evidence.
Another challenge was getting multiple AI agents to work together. We had to make sure that the output from the AI agents was consistent, structured and explainable.
We had to ensure that the recommendations from the AI agent were different to the actual payment authorisation. We solved this by creating a deterministic Payment Policy Engine.
Creating payment integration was another challenge we faced. We had to ensure that there was no confusion between simulated transaction, sandbox responses and real payment statuses.
This challenged us to think outside of creating an AI application. We focused on creating a reliable financial intelligence platform.
Accomplishments that we're proud of
We are extremely proud of our development of PayPal 2.0 as an Intelligent Trust Layer built using AI to help improve, secure and make transactions in the digital world smarter.
A major achievement we’re very proud of was creating a solution that uses various AIs to look at different aspects of trust. We have many AI’s looking at each aspect to give multiple opinions and ensure we have an adversarial AI to question and find any holes.
Overall we are very proud of how we’ve focused the product on providing explainable AI. Rather than just saying if it’s a safe or risky transaction, we want to provide clarity on why we’ve made those decisions, what information is missing and recommend what security measures should be put in place.
An important aspect we’ve focused on with our product is ensuring that AI is only used to inform financial decisions. By combining the AI intelligence with deterministic payment rules we can create a more responsible AI solution.
What we learned
Developing PayPal 2.0 taught us a lot about how to integrate an AI solution.
We learned how to break down complex problems using multi-agent AI architecture. We made multiple agents that work together to analyse data and provide different conclusions.
We also learned that trust can be broken down into more parts. We had to take into account factors such as identity verification, transaction risk, business trustworthiness and confidence in data.
We also learned the importance of responsible AI when it comes to financial solutions. All recommendations must be explainable, missing data shouldn’t be assumed to be fraudulent and payment authorisation must be kept secure with deterministic methods.
Using PayPal’s payment APIs helped us understand the payment lifecycle, sandbox, API integration and the importance of secure transactions.
What's next for PayPal2.0
We aim to create a dynamic trust intelligence system for PayPal 2.0. Our goals are:
Advanced Identity Intelligence – Incorporate trusted identity verification providers to improve our understanding of counterparties AI Agent Reputation – Create a reputation system to help evaluate autonomous AI agents before any financial transactions are made. Real Time Risk Detection – Identify abnormal patterns and risk. Continuous Trust Assessment – Continue to assess trust based on new transaction & fulfillment data. Increased use of Paypal APIs – Use more of the PayPal APIs to create secure payments, authorise payments, track transactions and receive webhooks. Explainable AI – Give more insight into the evidence, findings, risk factors and reasoning behind each recommendation. Using trust intelligence to scale – Consider using trust intelligence to assist individuals, businesses, merchants and AI driven commerce platforms.
Ultimately we would like to create a trustworthy intelligence layer for PayPal 2.0 so people and AI agents can make more informed decisions prior to any monetary transactions.
Long Term Vision – Make trust the core of every digital transaction.
Built With
- ai
- fastapi
- frauddetection
- largelanguagemodel
- multi-agent
- paypalapi
- paypalsandbox
- python
- supabase
