Projects
RE Automated Valuation Model
Problem: Manual real estate property appraisal takes an unnecessarily long time to produce, and excessive effort to make, it is also very expensive to order. This process is not scalable.
Solution: Using real estate transaction and property description data, we developed the best-in-market AI-driven algorithm that evaluates the prices of real estate properties with the highest accuracy.
Mortgage Underwriting Automation
(currently in POC with 2 banks).
Problem: It takes a few weeks to issue a mortgage. It relies on inaccurate collateral appraisal.
Solution: Using the client’s bank transactions data, users' personal data, and bank's underwriting rules (as well as our own property valuation model), we created a product that evaluates risks and completes all the steps of mortgage underwriting online within seconds.
Dynamic in-app content generation
Problem: Our clients who run interactive mobile applications or games always search for ways to improve engagement and per-user monetization.
​
Solution: we've built an ML model that identifies the moment when the user is most likely to take an action (make a purchase or drop off) and then feeds it into the content generation mechanism to create incentives for a purchase or retain the user.
PLTV calculation for CAC optimization
Problem: Digital product operators seek ways to reduce CAC.
​
Solution: We've built an ML algorithm that predicts LTV for users, acquired through different channels and adjusts the target CA spend for each channel. This can also be paired with an attribution mechanism, that identifies the user's engagement (and corresponding costs) in earlier stages of the sales funnel.
Risk monitoring for a financial regulator
Problem: People take out mortgages that impose unnecessary risk on themselves and the financial system as a whole.
​
Solution: We developed an algorithm that analyzes bank transactions of mortgage holders, and indicates potential risks, reports them to the regulator, together with the expected non-payment timeline (based on the loan equity evaluation)
Employees' office needs evaluation
Problem: Employers need to understand patterns of office usage.
​
Solution: Using employees' location data, and office usage data, we created a machine learning algorithm that forecasts office usage patterns and shows a footprint that helps employers find optimal workspace solutions.
Leasing risk assessment
Problem: A residential rental property management company needs to evaluate which tenants carry the lowest delinquency risk.
​
Solution: We’ve built a machine learning algorithm that analyzes applicants' bank transaction data and identifies spending patterns that indicate the true level of income and ability to pay the rent.
Recommendations mechanism for an online school
Problem: An online school needs to increase student enrolment and engagement
​
Solution: We’ve built ML algorithms that personalize learning experiences through customized content recommendations that target students' individual needs and interests. Meanwhile, the same model assesses the probability of each student group purchasing it, which helps with future content creation.
Data gathering for RE analytics company
Problem: Insights need to be automatically extracted from real estate discussions on YouTube.
​
Solution: We’ve created a scalable, reliable, and cost-efficient system for extracting and analyzing sentiment from real estate discussions on YouTube and turning that into structured data for future use.
