FairMoney

AUC is useful. But AUC does not approve a customer, set a limit, price a loan, choose a tenor, or manage exposure. Decision logic does. I have seen this debate come up often in lending: A new model is built. The AUC improves. The team gets excited. Fair enough — better rank ordering matters. But then comes the harder question: So what decision actually changes? Because a model can be statistically better and still create limited business value if: ✅ the lift is in score ranges the business does not use ✅ calibration is unstable ✅ cut-offs stay the same ✅ pricing and limits do not respond ✅ product constraints override the model ✅ implementation logic changes the intended decision That is where model performance and credit strategy need to meet properly. This is not an anti-model view. It is the opposite. Good models matter enormously. But in lending, model lift only becomes valuable when it improves the actual decisions the business makes. The real question is not only: “Did the model improve?” It is: “Did the lending decision improve?” Because at the end of the day, the book does not perform on AUC. It performs on the quality of decisions made with the model. #CreditRisk #Decisioning #LendingStrategy #ModelRisk #PortfolioManagement

Excited to announce that it’s been exactly 6 months since I commenced my MBA program with Nexford University and I must say it’s been the most challenging journey I have ever embarked on, especially with having to combine it with my 9 to 5 career. Someone might ask, How have you been navigating? Balancing a full-time 9-to-5 with rigorous academics requires immense discipline-waking up at 3:00 AM every day to study before work is the only way I make it happen. It’s been a really tough journey but seeing my grades at the end of every course makes it all worth it!! In these 6 months journey do far, I have successfully completed 6 courses managing to pull off straight A’s; Six more to go! Cheers 🥂 to all Professionals working hard to level up! You are doing amazing… #careergrowth #MBA #Nexford

After internally contemplating whether or not to drop this publicly, since it was supposed to be some sort of personal experiment, I am sharing 𝗣𝗲𝗮𝗿𝗕𝗼𝗼𝗸, a Proof of Concept (PoC) project built to validate the findings in my research paper, "𝗕𝗮𝗹𝗮𝗻𝗰𝗶𝗻𝗴 𝗖𝗔𝗣: 𝗔𝗰𝗵𝗶𝗲𝘃𝗶𝗻𝗴 𝗘𝘃𝗲𝗻𝘁𝘂𝗮𝗹 𝗖𝗼𝗻𝘀𝗶𝘀𝘁𝗲𝗻𝗰𝘆 𝗶𝗻 𝗗𝗶𝘀𝘁𝗿𝗶𝗯𝘂𝘁𝗲𝗱 𝗦𝘆𝘀𝘁𝗲𝗺𝘀 𝘂𝘀𝗶𝗻𝗴 𝗖𝗥𝗗𝗧𝘀 𝗮𝗻𝗱 𝗞𝗮𝗱𝗲𝗺𝗹𝗶𝗮 𝗗𝗛𝗧." 𝗣𝗲𝗮𝗿𝗕𝗼𝗼𝗸 is a decentralized, peer-to-peer expense tracker designed to demonstrate how we can maintain data integrity and consistency in a distributed network without a central server. By marrying Conflict-free Replicated Data Types (𝗖𝗥𝗗𝗧𝘀) with a Kademlia Distributed Hash Table (𝗞-𝗗𝗛𝗧), the project explores practical solutions to the trade-offs described in the CAP theorem. Despite the challenges of balancing this with my primary day-to-day work, I've managed to hit several key milestones on this PoC: • Core Architecture: Implemented a real Kademlia DHT using libp2p for decentralized storage and node discovery. • Consistency Layer: Developed custom CRDT implementations, including OR-Set for membership, OR-Map for expenses, and PN-Counters for balance tracking. • Propagation: Integrated Gossip protocols to ensure efficient data dissemination and state synchronization across the network. • Security: Integrated client-side ECDSA cryptographic signing for all peer operations. • Performance: Built a 16-shard local cache and worker-based background syncing to ensure high concurrency. • CLI & API: Developed a robust Go-based CLI and RESTful API for seamless node interaction. 𝗪𝗵𝗮𝘁'𝘀 𝗡𝗲𝘅𝘁? While the current PoC is built in Go, I am planning a major evolution. I intend to rebuild the project using the Holepunch 𝗣𝗲𝗮𝗿 𝘀𝘁𝗮𝗰𝗸. Leveraging Pear's P2P stable runtime and the Hypercore protocol will allow for a truly unstoppable, zero-infrastructure application that can be deployed seamlessly across Desktop and Mobile. I'm open for discussions and would love to connect with anyone interested in distributed systems, P2P networking, or the Holepunch ecosystem. I'd especially love to see people building applications marrying Kademlia DHT and CRDTs in creative ways! Paolo A. Check out the project and the full research paper below: 📂 GitHub Repo: https://lnkd.in/ef4TJiAf 📖 Research Paper: https://lnkd.in/eg4RPEGy #DistributedSystems #P2P #OpenSource #Holepunch #PearStack #Golang #CRDT #SoftwareEngineering #Research #CAPTheorem #Decentralized

A strong model is not the same as a strong decisioning system. A model predicts. A decisioning system turns prediction into action: ✅ who to approve ✅ how much to lend ✅ what tenor to offer ✅ what price to charge ✅ how to manage exposure over time That distinction matters. Many lenders improve models while the actual decision system remains fragmented across rules, overlays, product constraints, and operational workarounds. Prediction matters. But in lending, value is created when good prediction is translated into good decisions. #CreditRisk #UnsecuredLending #Decisioning