Data Cowboys — Work & Publications

Current Work

LLM Chatbots for Youth Well-Being

Sergey is Head of AI at Alongside, a clinician-developed youth wellbeing coach that helps students navigate school stress, friend drama, and big emotions through brief, guided chats and skill-building plans. As Head of AI, he leads the design of structured (non-companion) LLM coaching, including topic modules (40+), multilingual support, and safety systems that detect serious concerns and route alerts to human support via school protocols. Alongside reports measurable impact in partner settings (e.g., reductions in anxiety and suicidal ideation; attendance improvements) backed by external studies and evidence standards. Ongoing since: April, 2022.

AI for Scientists

Sergey is a Principal Research Scientist at Ai2, building production-grade AI systems for scientific discovery and knowledge access. His work sits at the intersection of information retrieval, entity resolution, and LLM/RAG evaluation, powering tools used by Semantic Scholar (200M+ papers) and Ai2's emerging Asta agent ecosystem. Ongoing since: March, 2016.

Contributed to Ai2 ScholarQA, an open RAG system for scientific QA and literature review with attribution; released as an open-source library.
Learning-to-rank search for large scholarly corpora. Overhauled Semantic Scholar’s search reranker using real search logs and released s2search alongside a detailed engineering tell-all.
Author identity resolution (S2AND) — released code, datasets, and reference models.
Built core components of a low-error paper clustering model (S2APLER).
Co-authored SPECTER (ACL 2020) and contributed to SPECTER2 (ACL 2023).
A paper about gender bias in clinical trial recruitment published in JAMA Network Open.
Blog post analyzing the association between early arXiv posting and citations.

Open Source Contributions

We contribute to the Python data science ecosystem. Sergey co-wrote the imputation package fancyimpute, and merged IterativeImputer into scikit-learn. Other packages we’ve worked on include:

Partnerships

Preva Group helps organizations achieve large-scale social change by combining structured and unstructured data with sophisticated analytics and ML. Data Cowboys operates a strategic employee-sharing partnership with Preva Group.

Previous Work

Machine Learning Strategy Consulting

The Healthy Birth, Growth, and Development (HBGD) program was launched in 2013 by the Gates Foundation. The Knowledge Integration (Ki) initiative aims to facilitate collaboration between researchers, quantitative experts, and policy makers in fields related to HBGD. The broad goal is to aggregate data from past longitudinal studies about pathways and risk factors that affect birth, growth, and neurocognitive development in order to better predict Ki outcomes. Sergey worked closely with Ki leadership—designing and overseeing data science contests, managing external collaborations with academic research labs and software companies, and modeling diverse global health datasets (an example is described here).

Improving Reading Comprehension

Actively Learn makes a reading tool that enables teachers to guide, monitor, and improve student learning. With our help, they wrote and were awarded an NSF SBIR grant to answer the key question: "How can we personalize reading instruction so as to increase comprehension & learning…" We are diving deep into the data with sophisticated machine learning tools, and bringing back testable hypotheses about what helps and hinders students.

Contributing to Technical Books

Jenny Dearborn, Chief Learning Officer and Senior Vice President at SAP, has written Data Driven and The Data Driven Leader. We helped her team communicate the mathematical models that are at the core of the book and contributed to plot and characterizations.

Recommendation Systems

Long Tail NLP-Based Recommendations. Most e-commerce recommendation engines have difficulty highlighting less frequently bought products. We developed a language-based model for RichRelevance that identifies recommendations based on product descriptions rather than purchase data.

Bayesian A/B Testing. Sergey built a comprehensive Bayesian A/B testing system to answer questions like “What is the probability that algorithm A has at least 5% lift over algorithm B…” Read the archived posts: [1], [2], [3].

Bandits for Online Recommendations. We engineered a bandit-based approach that learns from less data and adapts to any optimization metric. Archived posts: [1], [2], [3].