Michael Yang

Senior Data Scientist / ML Engineer driving revenue impact through predictive modeling and experimentation.

Princeton, New Jersey, United States

About

Dun and Bradstreet - Senior/Lead Data Scientist — Agentic Research & AI Team Architected and deployed distributed systems to process, chunk, and index massive volumes of corporate financial statements into a highly scalable vector search database, enabling sub-second semantic retrieval across millions of corporate entities. Engineered end-to-end RAG pipelines utilizing LangChain and Llama LLMs to extract complex financial insights, successfully capturing granular organizational metrics such as localized sales figures and dynamic employee-to-revenue ratios. Designed and implemented multi-agent workflows featuring advanced autonomous routing and state management to automate deep-dive financial research and compliance auditing. Optimized prompt engineering frameworks alongside rigorous evaluation methods (e.g., Ragas, LLM-as-a-judge) to systematically mitigate hallucination risks, track faithfulness, and maximize retrieval precision. Drove inference optimization strategies to minimize latency and compute overhead during high-throughput token generation, ensuring seamless downstream API integration with core enterprise product ecosystems.

Experience

  • Data Scientist at Dun & Bradstreet
    Jan 2020 - Present · 6 yrs 7 mos

    Architected and deployed distributed systems to process, chunk, and index massive volumes of corporate financial statements into a highly scalable vector search database, enabling sub-second semantic retrieval across millions of corporate entities. Engineered end-to-end RAG pipelines utilizing LangChain and Llama LLMs to extract complex financial insights, successfully capturing granular organizational metrics such as localized sales figures and dynamic employee-to-revenue ratios. Designed and implemented multi-agent workflows featuring advanced autonomous routing and state management to automate deep-dive financial research and compliance auditing. Optimized prompt engineering frameworks alongside rigorous evaluation methods (e.g., Ragas, LLM-as-a-judge) to systematically mitigate hallucination risks, track faithfulness, and maximize retrieval precision. Drove inference optimization strategies to minimize latency and compute overhead during high-throughput token generation, ensuring seamless downstream API integration with core enterprise product ecosystems. Re-wrote from scratch employee predictions engine for central Sales & Employee product ($60M) that supports a family of products ($1B). Factor analysis and selection using advanced AI methods and CatBoost regressor. Modified product's underlying objective function to avoid underestimation by deforming gradients and hessians in original model. Lead Data Scientist on a company-wide reaching new Global Rollup product to replace existing rollup (hierarchical business reconciliation algorithm) and reconciliation engines, useful for auditing large datasets. Implementing a LP-Solver based solution for a collection of roughly 7 million global family tree members. After one full year of model testing in 2021 assessing robustness, followed with a full Implementation into the US and International production pipeline during 2022 Q1-Q2.

  • Developer at Spindrop
    2012 - Jun 2023 · 11 yrs 6 mos

    Founded company and built patent-protected fully-automated AI Robotic DJ algorithm. Currently 4K+ monthly active users and 400 monthly organic users.

  • Data Scientist at Defense Advanced Research Projects Agency (DARPA)
    Jun 2018 - Nov 2019 · 1 yr 6 mos

    DARPA project through Culmen International using Twitter to predict human protest events in Russia, Venezuela, and Mexico to supplement on-phone IDFA location data. AI/ML statistical model authored could successfully predict protest activity 48 hours in advance using indicators and natural language processing.

  • Data Scientist at Ozone.ai
    Jan 2019 - Jul 2019 · 7 mos

    Flexible AI/ML solutions for data piping and advanced content delivery. Patented file chunking and transfer algorithms compatible with Blockchain. Wrote proprietary solutions for physical security risk (buildings, natural disasters, flood, crime), using public and proprietary big data.

  • Data Scientist at Findyr
    Jun 2018 - Jan 2019 · 8 mos

    Designed and implemented AI/ML forecasting models for insurable tail-risks SaaS product, focusing on perils arising from extreme weather, cyber incidents, and political violence. Led engineering team efforts for new product innovations in hospitality guest safety, market research, and and programmatic surveys.