andrew@danks.me
Based in San Francisco, CA
Citizen of Canada 🍁
Andrew Danks
Backend engineer with wide range of experience in distributed web applications, databases, stream processing and machine learning
Senior Software Engineer Yelp
Search & Recommendations
Search Suggest Lead engineer for the autocomplete system; rearchitecting the ranking engine with a new ML platform; improving performance and infrastructure
Filter Recommendation Led 4 engineers to make suggested search filters more contextual to the user's query, time of day, etc. with xgboost-trained classifier
User Location Intelligence
Realtime Store Visits for Flink application to classify customer visits from background user location pings streamed from Kafka. Implemented async i/o and improved throughput by over 1000 messages/sec
Data Quality
Chain Detection Led project to rearchitect system to detect chains (like Starbucks, In-N-Out, etc) at scale with Spark and machine learning. Presented at PyBay.
Location Quality Classifier Trained a model that detects businesses with inaccruate geocodes via user location data. Optimized decision threshold for different use cases. Created a feedback loop that solicits users and moderators to correct such locations.
Biz Gems Mined review text and developed classifier to infer features of a business such as "Good Date Spot", "Good for Groups", has "Outdoor Seating"
Business Coverage Metrics Developed an extendable system to calculate the coverage of business features (such as % of eligible businesses with Has WiFi value) aggregated on city and category and weighted by page views, all in a daily Spark job.
Business Match Owned critical data ingestion system powered by machine learning classifier with ElasticSearch as recall engine. Improved NLP for name and address comparison for better precision & recall.
Site-wide business data cache Led effort to reduce load on DBs for business-related data by over 40,000 QPS with 95% mean hit rate via Memcached.
Software Engineer Intern
Yelp
Business duplicate detection and search with ElasticSearch
Software Engineer Intern
Marin Software
Software Engineer Intern
ALM, Inc.
Research Assistant
Computational Linguistics Group
Developed acoustic and lexical features to classify
trouble-indicating behavior in Alzheimer’s patients
Teaching Assistant
Engineering Large Software Systems
Neural network library for Julia supporting backpropagation, CNNs, stacked denoising autoencoders, dropout, and more.
WiggleMap
Originated as the 1st Place winner ($7500) from the Google sf/reroute hackathon, a Ruby app that maps safe bike routes in San Francisco using crime, accident, and elevation data.
Python, Java, Scala, MySQL, Spark, Kafka, Flink, machine learning, ElasticSearch