Dec 2022 – Present · 3 yrs 5 mos · Berlin, Germany · Hybrid

• Making search go brr with ML:
• Making food search work better: Elasticsearch relevance, NLP, dense retrieval with embeddings, LLMs, fine-tuning with SBERT/PyTorch. Once built an LLM which can generate a dad joke punchline on any prompt.
• Convincing team and management to do the right thing: writing RFC/PRD/OKR/WTF docs after figuring out what the actual customer problem we are facing, and how can it be solved step-by-step. Hands-on development with MVPs, prototypes, and the most challenging parts of the system.
• Mentorship and team growth: organizing team/company hackathons, architecture review, paper reading club, and having 1-1 && pair-prog sessions about technical and ML excellence. Convincing colleagues to speak at conferences/meetups.
• Moving industry forward: open-source (metarank, lightgbm4j, flink-scala), speaking on tech conferences/meetups, flame wars on linkedin/hackernews/reddit/slack. Editor-in-chief of a #memes channel on a company Slack.

Jun 2022 – Present · 3 yrs 11 mos · Berlin, Germany

• Pet project which has grown into a full-featured open-source Learn-to-Rank service.
• Designed and implemented an open-source LTR engine with 2k+ github stars [Scala/Java, Kubernetes, Kafka/Kinesis/Pulsar, http4s, FS2, Cats-Effect].
• Technical marketing: blog posts (top-HN #2), community management (300 members in slack), integration guides for open-source search engines.
• Solution engineering for early adopters, conference speaking (MICES, Haystack, FlinkForward)

Mar 2016 – May 2022 · 6 yrs 2 mos · Berlin Area, Germany

• Leading a small team building smart search and recommendations for 2K+ merchants.
• #2 employee, designed a ML ranking system helping company grow from 30 to 2000 e-commerce stores [LambdaMART, XGBoost, Apache Flink, Cassandra, Redis].
• Migrated the offline LTR backend to online LTR, improving CTR +4% and CR +6% across the whole merchant base [Apache Flink, Cassandra].
• Led the migration from ElasticSearch to a custom Lucene-based cloud-native & multi-tenant search backend, reducing AWS costs 2x [Scala, CE3, FS2, Kubernetes, Cassandra, Apache Pulsar]
• Built an internal recommendation system for 20M products [MF/BPR, Java, HNSW/Annoy]
• Pitched a pilot project to migrate to k8s [v1.6, 2017], and led the complete company migration into it afterwards [v1.8, 2018]. Introduced GitOps/IaaS approach with Terraform and ArgoCD.
• Led a small team working on a real-time search analytics backend: sub-second dashboard update latency for 2k merchants and 80M MAU [Apache Flink, Clickhouse]
• Migrated high-thoughput (50gb/day) logging stack from cloud SaaS solution to Grafana Loki, cutting costs 4X.

May 2012 – Feb 2016 · 3 yrs 9 mos · Voronezh Region, Russian Federation

• Designed and implemented a major part of the analytical pipeline from the ground up having limited resources:
• Distributed large-scale web scraping (C++, Scala+akka, ~1.5b requests/month/server)
• Data processing: creating automated data extraction system based on Spark/Cassandra and Hadoop/HBase.
• Solving partial duplicate detection tasks in large databases.
• Creating advanced client credit scoring models for banking and finance (Weka, R, Rapidminer).
• Building fault-tolerant high-performance client API (Scala+spray, Cassandra, Elasticsearch)
• Keywords: Hadoop, HBase, Cassandra, ElasticSearch, Solr, Puppet, Zookeeper, Giraph, Java, Scala, Jenkins, Maven, Zabbix, RapidMiner.

Nov 2010 – May 2012 · 1 yr 6 mos

• Datamining, OLAP: creating client segmentation schemes, client scoring.
• Business research and modeling: building recommender systems.
• Social networks: advanced statistical research, building automated web crawlers/spiders, deploying large NoSQL databases.

Jan 2009 – Nov 2010 · 1 yr 10 mos

• Algorithmic trading research in option pricing. Creating and testing financial models in Mathematica & C++.
• High-volume trading robot research.
• Design and prototyping of trading robots.