Residential Performance Model (RPM)

Residential Performance Model (RPM)

Residential Performance Model (RPM)

Residential Performance Model (RPM)

AiCurio Mortgage Score (AiCR)

AiCurio Mortgage Score (AiCR)

AiCurio Mortgage Score (AiCR)

Servicing Optimization and Performance Improvement (SOAPi)

Servicing Optimization and Performance Improvement (SOAPi)

Servicing Optimization and Performance Improvement (SOAPi)

Servicing Optimization and Performance Improvement (SOAPi)

SECURED PARTY SALE BY PUBLIC TIMED ONLINE AUCTION 26-22
**************************************************************************
AI-powered Mortgage Analytics Platform - Software and Modeling Modules
**************************************************************************
RE: AiCurio
**************************************************************************
ONLINE BIDDING BEGINS: SUNDAY, MARCH 1, 2026 AT 9AM
BIDDING ENDS: MONDAY, MARCH 16, 2026 AT 1PM
**************************************************************************
TO VIEW ONLINE BIDDING SITE AT PROXIBID.COM AND REGISTER TO BID CLICK BELOW:


ABOUT AICURIO:
AiCurio specializes in artificial intelligence-powered mortgage analytics within the financial services industry. It offers a solution that enables credit institutions and mortgage professionals to predict risks, profitability, and cash flows. It uses AI-powered machine learning, deep learning, and artificial neural network for operations.

ASSET INFORMATION PACKAGE: Please Click Here to View Google Doc File or download from the above Proxibid link.

Residential Performance Model (RPM) - The Model evaluates the probability of a mortgage transitioning between statuses based upon the origination characteristics, performance history, mortgage product characteristics, and expected market conditions. The model was created using publicly available loan-level non-agency, Home Affordability Refinance Program (“HARP”), Fannie Mae (“FNMA”), Freddie Mac (“FHLMC”), and Ginnie Mae (“GNMA”) data, as well as privately held portfolio loans. The model output is expressed as a probability of the mortgage transitioning from one of six active statuses to nine ending statuses. The transition probabilities are estimated by a deep learning artificial neural network (“ANN”) model on the Momentum platform of accure.ai.

AiCurio Mortgage Score (AiCR) - This is a mortgage underwriting score based upon the expected profitability of the loan. AiCR is based on a series of deep artificial neural networks that have been trained with an unprecedented amount of mortgage origination and performance history spanning over two decades. The deep learning model found risk layering and compensating factor relationships that influence borrower behavior stemming from the borrower, loan, property, and macroeconomic inputs. The results of this model have significant implications for the mortgage market. The rank ordering of mortgage profitability can offer invaluable insight to risk exposure, and the need to diversify a portfolio across mortgage characteristics and geographic location. This insight can lead to better investment decisions, pricing, ratings, and hedging of whole loans. Additionally, AiCR can offer a consistent mechanism to safely expand the credit box to offer mortgage financing to viable underserved borrowers.

Servicing Optimization and Performance Improvement (SOAPi) - The engine powering retention, collections, and loss mitigation guidance. Simulation of different actions available for high-risk loans. Generate new performance and cash flow forecasts for each potential action. Determine the net present value (NPV) gain associated to execution. Provide a list of beneficial next steps.

Explanation of Servers:

- Fortinet Firewalls. Protection against outside network attacks. These are well-known and used by many large companies. While they currently have basic configurations, they have the capability of being very robust and meeting enterprise compliance measures.

- Research Database Server. Collection and preparation for model training. We collect and prepare 4 public datasets (Fannie, Freddie, Ginnie, and CTSLink) on a quarterly basis. This equates to roughly a little over 4+TB of data and growing each quarter.

- Model Training Server. Independent server using Momentum. We train the models using Accure's Momentum application.

- Main/DW/Report Database Server. Data repository of everything non-historical related. This includes data for clients, metadata (for mappings, translations, and imports), and results from model valuations. This server can also store a large amount of data.

- (2) Model Valuation Servers. 2-redundant servers with GPUs for a time-series loop that use the 6 models to produce loan transition probabilities integral in calculating cashflows' for each time-period evaluated. It is best to have at least 2 model valuation servers because when we re-train the model to the next version, we consequently need to allow users access to the old version. These 2 servers allow us to transition.

- Web Server/Third-party Factors. Front-end server that allows users to access the Main/DW Database and Report Server. It is the only server that has access to outside the network, minimizing the risks of hacks. Also serves as a daily collection of economic and other third-party factors used in the model valuations.

SOFTWARE HOSTING/SERVER RENTAL: AiCurio software and modules are currently stored on multiple servers at 400 West Cummings Park, Woburn, MA. Winning bidder will be required to make arrangements to continue use at the current data center or alternate server plans. Additional information is provided in the Asset Information Package that will be made available on the online bidding site at Proxibid.com on or before March 1, 2026.

TERMS: Payment in US FUNDS by Certified Check or Wire Transfer. 5% Buyer’s Premium. For more information contact Auctioneer's office at (207) 885-5100 and request auction #26-22, Stefan P. Keenan, MA License #1881. Our 54th Year & 8,860th Auction.