Craigslist Semantic Search Engine
Semantic search engine to match Craigslist ads by intent rather than keywords
Situation
Craigslist's keyword-based search was inefficient, often missing relevant ads due to synonym mismatches and poor query understanding.
Task
Create semantic search engine that matches ads by intent and meaning rather than exact keyword matches.
Action
Business & Strategy
Redesigned user journey for ad discovery, incorporating relevance feedback loops. Conducted user testing to validate search quality improvements.
Technical Implementation
Implemented TF-IDF vectorization and cosine similarity for baseline. Fine-tuned BERT embeddings on Craigslist-specific corpus for semantic matching. Built REST API with Flask for search queries. Optimized query latency to <200ms using vector indexing.
Results
Business Impact
Relevant match rate increased by 27%. User engagement time increased by 34%.
Technical Achievement
F1 score of 0.89 after BERT fine-tuning. Successfully processed 10,000+ queries/day with 99.9% uptime.