How I work
Two roles — open one
My work runs on two tracks. As a Professor at the University of Michigan, I build child-welfare data infrastructure and analytics for state agencies and public-sector partners through the Child & Adolescent Data Lab. As an advisor through Parallel 42, LLC, I provide fractional AI advisory services to private and mission-driven organizations.
NLP & machine learning for administrative data
Domain-adapted encoder models (including CPS-BERT, built on ModernBERT) and construct classifiers for high-stakes text; detection of substance use, domestic violence, firearm presence, opioids, and housing instability within case records.
Privacy-preserving & local AI
De-identification pipelines for sensitive narratives using surrogate-replacement methods and span-based PII detection; fully local, air-gapped model deployment that avoids third-party cloud processing.
Model fine-tuning & adaptation
Parameter-efficient fine-tuning (LoRA and QLoRA), Unsloth-based training, multi-GPU workflows, and teacher–student active learning pipelines.
Data engineering & analytic infrastructure
High-throughput analytic pipelines built on DuckDB over Parquet; multi-phase, reproducible data-asset architectures; large-scale corpus construction and vector search.
Applied AI development
Retrieval-augmented generation systems, agentic workflow harnesses, and web-based dashboards and decision-support applications.
Legal & policy AI
AI-assisted document review and drafting for legal contexts; document-archive and brief-preparation systems operating in secured environments; AI literacy leadership within a law school clinic.
Model benchmarking & evaluation
Systematic benchmarking of encoder models against generative large language models; embedding-based retrieval evaluation; scientometric analysis of AI-assisted writing.
Data visualization & performance dashboards
Architecture and development of operational dashboards for public-sector decision-making.
Training & capacity building
Curriculum design and professional-education delivery on responsible AI computing; doctoral-level instruction in computational text analysis; workshops for practitioner and academic audiences.