video1.0<iframe src="https://www.loom.com/embed/4158cfcfc12c41a4b3bffc0ce2a10020" frameborder="0" width="1920" height="1440" webkitallowfullscreen mozallowfullscreen allowfullscreen></iframe>14401920Loomhttps://www.loom.com14401920https://cdn.loom.com/sessions/thumbnails/4158cfcfc12c41a4b3bffc0ce2a10020-04ed65f1ac938913.gif298.731Hi guys, I wanted to share a proof of concept solution for a real estate company and the lessons I learned from it. We only had a little over 2,000 transactions to train a model to predict property price, and the simulations did not fully mimic real life due to poor location data, unrealistic temporal patterns, and asset value issues. Because asset value was not available for customer searches, I built two models, a fallback model without asset value for customers and an assessment aware one for agents to reduce workload. The agents can also see similarity property ranking for nearby price matches. No specific action was requested, I just wanted to document what we would need to improve for large scale.