Jesse Windle

DYDX Capital→

Architected and built an end-to-end data-driven deal sourcing platform that processes thousands of early-stage companies per month and reduces the set requiring human review by approximately 10x.

• Designed a three-stage funnel pipeline — regex rules (geography, industry), LLM agent (company-fit scoring), and ML quality model — that filters thousands of companies down to a curated daily shortlist.
• Built an ML model predicting company quality from founder demographics (e.g. work history and education); founder features significantly improved average precision over the company-demographics-only baseline.
• Developed a unified model infrastructure wrapping PyTorch Lightning and scikit-learn under a single Python and CLI fit/test/predict API, enabling systematic evaluation of 3+ model families (penalized linear, gradient boosted trees, neural networks).
• Built a separate model comparison platform — database, REST API, and frontend — for evaluating models along performance metrics and exploring how modifications to models and features perturbed company quality rankings.
• Built a multi-source feature pipeline supporting SQL and JSON inputs from multiple data providers, with output formats ranging from one-hot encodings to sentence-transformer embeddings, allowing the quality scoring model to ingest heterogeneous founder and company data at production scale.
• Built the agentic pipeline stage using an LLM to assess company descriptions against the firm's investment thesis and return a continuous fit score with a natural-language rationale, replacing manual description review with automated, explainable filtering at scale.
• Built a web platform (Next.js, Prisma, Clerk, Vercel, GitHub Actions) giving partners a daily feed of personalized company recommendations, document access, AI-generated commentary, and review workflow.
• Built a schema-agnostic caching and retrieval system for web-based data sources (TypeScript, Next.js, Prisma); the general design accommodated arbitrary source schemas, controlled external data costs, and accelerated retrieval across all tracked companies.
• Independently scoped and executed a greenfield project from first principles — moving from an undefined problem to a production system — with the platform ultimately becoming a core part of the firm's daily investment process.
• Stack: Python (FastAPI, SQLAlchemy, Pandas, NumPy, scikit-learn, PyTorch, Lightning, statsmodels, sentence-transformers, cvxpy), TypeScript (Next.js, Prisma, Clerk), PostgreSQL (Neon), Vercel, AWS.

Self-employed

Early technical validation work for DYDX Capital.

• Conducted an extensive literature review of startup success prediction and VC deal sourcing methods, benchmarking existing datasets and reported model performance and identifying significant shortfalls in prior work that shaped our own approach.
• Completed a comprehensive model review establishing baselines, best-performing model families, and expected performance bounds, providing the empirical foundation that justified the modeling investment and guided architecture decisions in the subsequent full build.
• Presented findings at DYDX investor day, building LP confidene in the data-driven approach.

Hi Fidelity Genetics / Technologies

Joined as the first employee of an agricultural technology startup and wore every technical hat — CTO, product manager, program manager, and data scientist — while building and leading a 15-person cross-disciplinary team. Co-invented a novel in-field root growth measurement device (RootTracker) and delivered quantitative results to leading agrochemical companies.

• Conceived and co-invented RootTracker, a device for capturing full-season, high-throughput, time-lapsed in-field root system dynamics — a capability that did not previously exist at scale.
• Co-authored and won 2 competitive grants (ARPA-E, SBIR) plus 2 follow-on awards, raising $2M+ in non-dilutive funding that financed device development and the core team build-out; managed quarterly progress reporting to the granting agencies.
• Built and led a 15-person team spanning data science, hardware engineering, software engineering, biology, plant breeding, and agronomy. Managed with a flat, high-trust "lab" style: weekly all-hands, weekly 1:1s with team leads, and teams self-organizing work within a high-level waterfall outline.
• Guided development of the company's complete data pipeline: hardware → AWS S3 → PostgreSQL → Flask/SQLAlchemy REST API → Python analysis, enabling hundreds of field devices reporting at 5-minute intervals to store, process, and expose high-throughput sensor data for downstream analyses.
• Led delivery of the company's core commercial product — scientific reports and experimental results from field, greenhouse, and lab trials — to clients including Bayer, Corteva, and BASF and academic partners across multiple seasons; managed the full cycle from experimental design through deployment, live monitoring, and final presentation. Client relationships were a primary driver of the company's acquisition strategy.
• Developed a novel 3D Bayesian statistical framework for modeling crown root growth trajectories in monocots (Stan, mixed models, state-space models, Gaussian processes), yielding compelling visualizations and quantitative insights into root architecture that resulted in a pre-printed manuscript.
• Built the company's data science and analytical infrastructure from scratch — Python pipelines for ETL, feature extraction, EDA, and visualization; R for statistical analysis, hypothesis testing, and reproducible reporting — supporting all client deliverables, grant reporting, and internal R&D.
• Stack: Python (Flask, SQLAlchemy, NumPy, xarray, Pandas, SciPy, scikit-learn), R (lme4/lmer, spaMM, Stan), PostgreSQL, AWS S3.

Duke University

Postdoctoral research in Bayesian statistics with Sayan Mukherjee; began work in quantitative genetics that led to the Hi Fidelity role. Taught introductory statistics as Visiting Assistant Professor.

University of Texas at Austin

University of Nebraska - Lincoln

Leadership and communication ()

team building and management, cross-functional team leadership, grant writing, scientific and technical writing, client and stakeholder communication, experimental design and project planning

Statistics and machine learning (Expert)

regression and classification, clustering and dimensionality reduction, hypothesis testing, variable selection, penalized linear models (Lasso, Ridge, Bridge), mixed models / hierarchical models, state-space models, time series, Bayesian inference and MCMC, neural networks, gradient boosted trees, random forests, Gaussian processes, ordinal regression, data visualization and EDA

ML / data science tools (Expert)

Python, NumPy, Pandas, SciPy, xarray, scikit-learn, statsmodels, PyTorch, Lightning, sentence-transformers, cvxpy, R, lme4, spaMM, Stan

Systems and engineering tools (Expert)

SQL / PostgreSQL, FastAPI, Flask, SQLAlchemy, Next.js, Prisma, TypeScript, Git / GitHub, Docker / Podman, AWS (S3, EC2), Linux

LLM / agent tooling (Proficient)

LangChain, OpenAI API, prompt engineering, agentic pipelines, RAG

Capturing in-field root system dynamics with RootTracker

Plant Physiology • November 2021

J.J. Aguilar, M. Moore, L. Johnson, R.F. Greenhut, E. Rogers, D. Walker, F. O'Neil, J.L. Edwards, J. Thystrup, S. Farrow, J. Windle, P.N. Benfey. Plant Physiology, 187(3):1117–1130.

A tractable state-space model for symmetric positive-definite matrices

Bayesian Analysis • December 2014

J. Windle and C. Carvalho. Bayesian Analysis, 9(4):759–792.

The Bayesian Bridge

Journal of the Royal Statistical Society Series B • September 2014

N. Polson, J.G. Scott, and J. Windle. Journal of the Royal Statistical Society Series B, 76(4):713–733.

Bayesian Inference for Logistic Models Using Pólya-Gamma Latent Variables

Journal of the American Statistical Association • December 2013

N. Polson, J.G. Scott, and J. Windle. Journal of the American Statistical Association, 108(504):1339–1349.

rootmodel

R and Stan code supporting the manuscript "Inferring monocotyledon crown root trajectories from limited data." Bayesian model for 3D crown root growth.

R • Stan • root modeling

gmmfun

Python package for fitting distributions via their moment generating function using the generalized method of moments.

Python • statistics

ctgauss

Python package for sampling from a Gaussian random variable conditioned on a piecewise linear function via HMC.

Python • statistics

BayesLogit

R package for sampling from the family of Pólya-Gamma distributions, enabling exact Bayesian inference for logistic and negative-binomial models.

R • Bayesian statistics

Inferring monocotyledon crown root trajectories from limited data

Manuscript on root growth trajectory modeling methodology.

root modeling • statistics

Sampling from a Gaussian distribution conditioned on the level set of a piecewise affine, continuous function

Manuscript on HMC sampling within constrained Gaussian spaces.

statistics • sampling

Efficient Data Augmentation in Dynamic Models for Binary and Count Data

Manuscript on Bayesian analysis of time series with binary and count observations using Pólya-Gamma augmentation.

statistics • Bayesian methods

Forecasting High-Dimensional, Time-Varying Variance-Covariance Matrices

Ph.D. Thesis, University of Texas at Austin, 2013.

statistics • time series