This document describes building a real-time search suggestions system using open source tools like Elasticsearch, Hadoop, Redis, Flume, and Node.js. Logs are collected using Flume and processed through two MapReduce jobs - the first counts search terms by time bucket, and the second calculates scores and ranks terms. Results are stored in Redis with timeouts. Improvements discussed include using request diffs to update only changed results and adding more signals like click data to improve suggestions.