About
I am one of the co-founders and CTO of .txt. We
raised $11.9 million in 2024 to improve the reliability of LLMs through structured generation. Before launching .txt in 2023, I spent a decade building and leading data science and engineering teams at companies like Ampersand and Normal Computing.
My technical expertise lies at the intersection of data science, software development, and AI/ML engineering. I have both personally built and led the teams to design, from the ground up, production machine learning systems across industries as diverse as media, asset management, and container freight shipping. I have bootstrapped software engineering and data departments for organizations ranging from startups to large corporations.
Work Experience
Co-founder and CTO - .txt - (10/2023 - Present)
- Co-founded and raised $11.9M to build the first company focused on LLM structured generation, the building block of AI systems that produce clean, well-formatted data instead of raw text.
- Led development of the .txt API, licensable binary distribution, and AWS Marketplace deployable structured generation products.
- Personally hired or recruited the entire founding team of 7 people, then scaled from 7 to 17 engineers within one year.
VP of Engineering - Normal Computing - (04/2023 - 10/2023)
- Led engineering at a deep-tech AI startup focused on production-ready, reliable, generative AI workflows.
- Scaled engineering team from 1 to 7 engineers in fewer than 6 months.
- Released Outlines, a leading open source library for structured generation (12K+ stars).
Senior Director - Data Science & ML Engineering - AmpersandTV - (10/2019 - 01/2023)
- Built, from the ground up, distributed Data Science and ML Engineering teams of 8 engineers.
- Co-authored technical blog post with AWS publicizing the end-to-end ML system my team built that served over 1.4 trillion forward-looking impression estimates with sub-second response-time using Python, FastAPI, and ClickHouse.
- Reduced average error of viewership forecasts by 40% using more than 200,000 fully Bayesian Hidden Markov Models, and released the core estimation framework, PyMC3 HMM, as Ampersand’s first open source project.
President & Founder - Enplus Advisors - (06/2011 - 10/2019)
- Successfully completed engagements with clients ranging from one-person startups to large multi-nationals, specializing in problems at the intersection of data science and software engineering.
- Solving problems at the intersection of data science and software engineering
- Ranked Clutch.co #1 boutique Data Analytics and Big Data Firm and #9 overall
in 2019
- Assisted portfolio managers in researching, implementing, and trading alternative investment strategies, including Global Market Neutral, Global Macro, and Tax-Managed.
- Developed methods and tools for decomposing the effects of portfolio constraints on signal performance, allowing for attribution of over 80% of difference between theoretical and implemented portfolios.
Open Source
bootES: Calculate robust measures of effect sizes using the bootstrap.
backtest: The backtest package provides facilities for exploring
portfolio-based conjectures about financial instruments (stocks, bonds, swaps,
options, et cetera).
portfolio: Classes for analysing and implementing equity
portfolios, including routines for generating tradelists and calculating
exposures to user-specified risk factors.
Programming with Data: Go from beginner to
practitioner using Python and pandas to manipulate tabular data. Taught to
1,000s of students around the work and assumes no experience with pandas.
Publications
Jeffrey Enos, Daniel Gerlanc, Brandon Willard, Pierre-Yves Aquilanti, and Ala Abunijem. Bayesian ML Models at Scale with AWS Batch. AWS HPC Blog, June 14, 2022.
Kirby, K. N., & Gerlanc, D. (2017). Finding Bootstrap Confidence Intervals for
Effect Sizes with BootES. APS Observer, 30(3).
Iyengar A, Paulus JK, Gerlanc DJ, Maron JL. Detection and Potential Utility of
C-Reactive Protein (CRP) in Saliva of Neonates. Frontiers in Pediatrics,
November 2014.
Daniel Gerlanc and Kris Kirby, bootES: An R Package for Bootstrap Confidence
Intervals on Effect Sizes. Behavioral Research Methods, March 2013. (Preprint)
Kyle Campbell, Jeff Enos, Daniel Gerlanc, and David Kane.
Backtests. R News, 7(1):36-41, April 2007.
Podcasts
Live Teaching and Recorded Courses
- Programming with Data: Python and Pandas (Recorded Training). Pearson, 2020.
- Python and Dask: Scaling the Dataframe (Live Training on oreilly.com). Pearson, May 2020 – Present.
- Programming with Data: Advanced Python and Pandas (Live Training on oreilly.com). Pearson, May 2019 – Present.
- Programming with Data: Foundations of Python and Pandas (Live Training on oreilly.com). Pearson, September 2018 – Present.
Education
Williams College, B.A., 2007
- Majored in Comparative Literature
- Created an unofficial “Data Science” course of study by combining coursework in Computer Science, Statistics, and Economics
(c) 2025 Dan Gerlanc