Comps Engine

The Problem

Comparable-property analysis is the foundation of every real estate decision. The available comps tools either pulled from a single source (and missed half the market) or required hours of manual digging.

The Solution

A Go-based comps engine:

• Scrapes 4 real estate sources in parallel using a worker pool. Concurrency tuned to respect rate limits while keeping latency low.
• 99.8% data accuracy validated against ground-truth samples and refreshed continuously.
• LightFM hybrid recommender with geospatial proximity weighting and social-graph signals to rank "what's actually comparable" instead of naive distance matching.
• Reusable ETL framework so adding a new source is a small contract change, not a rewrite.

Why Go

The original implementation was sequential Python. A 40-minute job became 4 minutes after I rebuilt the orchestration in Go with proper worker pools, channels, and DynamoDB-backed state. Same workload, 10x throughput.

Outcome

Comps that used to take an investor an afternoon now run in seconds. The ranking quality is good enough that investors trust the top 5 results without manual review.