Benjamin Han

Leading AI/ML R&D @  | Agentic AI, Knowledge Graphs, LLMs

Seattle, Washington, United States

About

I lead AI/ML R&D at the intersection of agentic AI, knowledge graphs, continual learning systems, and LLMs, taking ideas from research to production. At Apple, I drive cross-functional efforts across AIML and Services Engineering, building mission-critical agentic systems and streaming knowledge graphs that support efficient, high-quality knowledge acquisition, reasoning, and adaptation over time. Previously, I led the Azure natural language science team at Microsoft, growing service usage by over 300% in two years and launching the first large-scale summarization service among major cloud providers. Earlier at IBM Research, I contributed to advances in information extraction, led award-winning entries in government-sponsored competitions, and transferred research into Watson products. I build teams and systems, from research programs and proofs of concept to robust, efficient production models that meet real-world SLOs. I hire and mentor ICs and tech leads, design evaluation rubrics that raise the quality bar, and enjoy working on problems spanning learning, reasoning, decision-making, and human-centered interaction. I am endlessly curious and enjoy building intelligent systems that can reason, act, adapt, and communicate with both rigor and empathy.

Experience

  • Apple (Seattle, Washington, United States)
    • Principal AI/ML Lead, Digital Supply Chain R&D,  Services Engineering
      Feb 2025 - Present · 1 yr 5 mos

      Leading AI/ML R&D in Apple Digital Supply Chain Engineering. Driving cross-functional initiatives in agentic AI, streaming knowledge graphs, continual learning systems, and hybrid reasoning architectures beyond LLMs.

    • Principal Scientist, Information Intelligence, AIML
      Sep 2022 - Feb 2025 · 2 yrs 6 mos

      Working on knowledge graphs, question answering, faithful generative AI, RAG, reasoning, and LLMs.

  • Microsoft (Full-time · 5 yrs 7 mos)
    • Principal Science Architect, Azure Cognitive Service for Language
      Jul 2022 - Sep 2022 · 3 mos

      Knowledge from and to natural language services!

    • Principal Science Manager of Azure Language Pillars (ALPS), Azure Cognitive Services
      Oct 2020 - Jul 2022 · 1 yr 10 mos

      Building and leading Science team dedicated to democratizing the state-of-the-art multilingual natural language research to serve customers at scale, including services for texts, documents, conversations, and transcripts. Guiding multiple engineering teams to ensure service quality and efficiency meets the highest standard. Grew usage of Azure Cognitive Service for Language 300+% over 2 years, and claimed the first in offering summarization services among the top cloud providers. Areas covered (all multilingual; visit language.azure.com for demos): * Key phrase extraction. * Named entity recognition, including both prebuilt and custom (allowing customers to label and train). * Entity linking. * PII redaction (both text and transcripts). * Sentiment analysis. * Opinion mining (aka aspect-based sentiment analysis). * Extractive summarization. * Abstractive summarization on long documents, contact center and meeting transcripts. * Language detection. * Text classification (allowing customers to label and train). * Text Analytics for health (NER, entity linking and relation extraction). * Question answering, including custom Q&A. * Conversational language understanding: intent classification and slot filling. * Relation extraction. * Coreference resolution. * Visual document understanding.

    • Principal Science Manager of Text Analytics, Azure Cognitive Services
      Dec 2019 - Oct 2020 · 11 mos

      Leading Science team dedicated to democratizing the state-of-the-art NLP research to serve customers at scale, including language detection, key phrase extraction, named entity recognition, entity linking, sentiment analysis, summarization, and more.

  • Program Committee Member at The Third Document Intelligence Workshop @ KDD2022
    Mar 2022 - Aug 2022 · 6 mos

    Organizing the 3rd Document Intelligence workshop with colleagues from Adobe, Google, IBM, and Redgrave Data.

  • Program Committee Chair at The Second Document Intelligence Workshop @ KDD2021
    Mar 2021 - Aug 2021 · 6 mos

    Successfully organized and chaired the one-day workshop on Document Intelligence with committee members from Google, IBM, Macquarie University, and Reveal-Brainspace, which includes 15 paper presentations reviewed by 40+ reviewers, and 6 invited speakers from Google, IBM, JHU, Microsoft, U. Michigan, and UIUC.

  • Research Staff Member at IBM
    2006 - Mar 2017 · 11 yrs 3 mos

    Member of Multilingual NLP Technologies group. Key contributor in the following selected projects: * IBM SIRE (Statistical Information and Relation Extraction): major contributor to IBM's comprehensive information extraction (IE) system; built time normalization engine, slot filler extractor and text region classifier, and have extensive experience in improving performance of mention detection, coreference resolution and relation extraction engines. * Watson Knowledge Studio (cloud-based IE suite): major contributor to the systems; built robust models for domains such as legal, business, life sciences, tech, finance, geology etc; organized company-wide training workshops for WKS and designed the certification process. * Watson for Cyber Security: NLP tech lead to extract vulnerabilities/indicators/threat actors from unstructured texts. * IBM and Nuance joint development project: developed mention detection and template-based extraction models on medical clinical notes to identify insurance billing codes (ICD-9/10). * Developed IE systems for many domains, such as extracting information on scientific papers (with machine-learned PDF layout recognition), identifying events for predicting regional political stability, and recognizing company acquisition events etc. * Team leader of the long-term annotation project KLUE (Knowledge from Language Understanding and Extraction) to produce high-quality multilingual data with rich semantics for building IE systems. * Participants to NIST-organized ACE and TAC-KBP competitions: 2nd place in ACE temporal expressions recognition/normalization task, and 1st-place in slot-filling task of TAC-KBP 2009 and 2nd in slot-filling of 2010. * Major contributors to various DARPA-sponsored projects (Machine Reading, GALE Distillation, etc): delivered domain-specific IE systems ranging from sports, intelligence to computational political science; built answer extraction models for template-based question-answering systems.