Evan Curtin
Principal Applied Scientist shipping LLM systems and evaluation in active litigation workflows
Brooklyn, NY | evanmcurtin@gmail.com | github.com/ecurtin2
Skills
ML: Applied LLMs & Agents, NLP, Evaluation (LLM + offline + production), Information retrieval (RAG), Active learning
Core: Python, SQL, NumPy/Pandas
Systems: Docker, Kubernetes, Spark, GitOps CI/CD, Terraform, Databricks, Prefect, AWS/Azure, Polars, Elasticsearch, C++/Fortran, Rust, Julia, DataFusion
Experience
Principal Applied Scientist - Relativity
- Shipped LLM-based document classification for eDiscovery (Relativity aiR for Review); exceeded contract-review performance (industry standard) and ran on dozens of millions of documents in active litigations.
- Built privacy-preserving evaluation pipelines and reliability metrics for LLM systems; enabled hundreds of experiments on sensitive legal data and partnered directly with AmLaw 100 firms (incl. partners) on active-litigation experimentation, enablement, and sales conversations.
- Built LLM workflows for damages assessment in high-value litigation; cut time per example from ~1 day to <2 hours and saved an estimated ~600,000 hours across a large law firm.
Data Engineer - Coalition, Inc
- Built CI for fast-changing insurance pricing models during hypergrowth; improved algorithm performance ~6× while reducing deployment risk.
- Scaled pricing systems with Dask and SageMaker; integrated real-time vulnerability data and deployed with Terraform.
Principal Machine Learning Scientist - Capital One
- Built daily-refit fraud detection pipelines processing >1TB/day; prevented ~$14M/year in fraud losses and deployed daily with >95% success in a regulated environment.
- Designed readiness checks (calibration + out-of-band validation) and GitOps CI/CD for fast iteration and safe rollback.
Data Science Engineer - Broadspire (Crawford & Company)
- Built XGBoost insurance risk models for adjuster triage; cut training time 90% with Dask and deployed via low-maintenance AWS Lambda services.
Education
PhD in Chemistry (ABD - All but Dissertation), University of Illinois, Urbana-Champaign
Focus: Stability of Hartree-Fock Equations in symmetry-breaking solids
Technologies: C++, SLEPc, Blue Waters supercomputer
MS and BS in Chemistry, Drexel University
Thesis: Low Dimensional Models for Predicting Nanomaterial Properties. Physics Minor.
Publications
- E. Yang, E. Curtin, et al., "Beyond the bar: generative AI as a transformative component in legal document review," 2024 IEEE Int. Conf. Big Data, 2024.
- G. Bazargan, E. Curtin, K. Sohlberg, "Comparing statistical predictions of quantum particle transit times in molecular systems to experimental measurements," J. Theor. Comput. Chemistry, vol. 18, no. 8, p. 1950039, 2019.
- E. Curtin, G. Bazargan, K. Sohlberg, "Quantifying electron transit in donor-bridge-acceptor systems using probabilistic confidence," J. Theor. Comput. Chemistry, vol. 17, no. 7, p. 1850046, 2018.
- E. M. Curtin, K. Sohlberg, "A reduced dimensionality model of torsional vibrations in star molecules," Physica E: Low-Dimensional Syst. Nanostructures, vol. 77, pp. 131-137, 2016.
- J.-C. Bradley, A. Lang, A. Williams, E. Curtin, "ONS open melting point collection," Nature Precedings, 2011.
Projects
- Quantized: An easy to use Quantum Mechanics Library in Python
- Cookiecutter-Python: A project template for python projects with CI and Codespaces support
- nbsanity: Jupyter notebook linter written in Rust
- mkdocs-apidoc: Plugin for using mkdocs with autogenerated API documentation
- Hartree-Fock Stability: Find if a HF solution is unstable with high performance linear algebra