About
Etymology
Sepanta — the name this garden grows under.
The Gardener
The hand behind Sepanta, and the spirit behind the tree.
I’m Kaveh Razavi — a Lead Staff Software Engineer working across data platforms, ML systems, and gen-AI. I’ve been building software for more than twenty years, the last twelve of them at Comscore. Most of my recent work there is data engineering at scale — Spark and Airflow pipelines on AWS — though my first Comscore years were spent on the mobile and cross-platform side. Before all that came graduate studies, research, and teaching at George Mason University — and, earlier still, several years of enterprise .NET work.
These days my work centers on large-scale data engineering — designing and running the pipelines that turn raw measurement into something products and teams can rely on, built to a simple standard: they hold up under real data, and their results can still be read and trusted a year later. What excites me most, though, is the architecture of gen-AI systems — retrieval, agents, and the platforms that make them dependable. A recent M.Eng. in Computer Science (Machine Learning) from Virginia Tech sharpened that pull, and it is where most of my open-source work now lives.
Sepanta is the Persian form of Spenta — the bounteous spirit in the verse above. I chose it as a commitment more than a name: build things that create, and tend what grows.
Experience
Comscore, Inc. — Lead Staff Software Engineer
2014 – Present
Back-end / Data Engineering
2019 – Present
Lead the design and modernization of large-scale audience-measurement platforms using Scala, Spark, Airflow, AWS, Snowflake, and Iceberg. My work spans connected television, streaming classification, identity resolution, privacy compliance, cross-media analytics, and reporting.
Defined and led migrations of legacy media-processing pipelines to cloud-native Scala/Spark workloads. Combined AI-assisted engineering with specification-driven development, automated testing, and rigorous data-parity validation to carry complex rewrites from discovery through production readiness.
Built and owned pipelines that transform panel, census, and streaming data into production measurement products. Translated evolving requirements into maintainable architectures, reusable processing stages, operational workflows, and documentation.
Developed privacy-conscious systems for data-subject request processing, identity resolution, enriched panel events, and privacy-preserving audience measurement. Led proof-of-concept work involving privacy-preserving cross-party analytics and statistical audience sketches.
Automated reporting and data-quality workflows and collaborated across engineering, product, analytics, and cloud teams to improve the reliability and maintainability of Comscore’s data estate.
Front-end / Mobile Engineering
2014 – 2019
Served as sole owner of Comscore’s .NET measurement SDK across 14 Windows, Windows Phone, Silverlight, WPF, and Xbox variants. Also owned its WinJS/Xbox, Xamarin, and Flash/ActionScript SDKs and contributed to Android, Apple-platform, and shared C++ implementations.
Led cross-platform feature alignment, releases, modernization, and long-term maintenance across the client-measurement surface. Contributed to consumer and panel applications spanning iOS, Android, Windows, and macOS while maintaining consistent measurement behavior across platforms.
Graduate Teaching Assistant & Directed Researcher — George Mason University
2011 – 2014
Taught engineering probability and statistics and introductory object-oriented programming with Java, designing coursework, evaluating student projects, supervising class sections, and holding weekly office hours. Conducted research in software architecture under uncertainty, co-authored two peer-reviewed papers, and developed the GuideArch decision-support tool.
Earlier engineering roles
2005 – 2010
Senior and lead positions across enterprise .NET at Parsaman Tarh Consultants, Sazeh Consultants, Rasa, and Idea Gostar IT: designing analysis, estimation, and reporting services over legacy platforms; building WPF and Silverlight MVVM applications backed by WCF and SQL Server; delivering reusable ASP.NET components for a model-driven content-management platform; and contributing cryptography and digital-signing capabilities to a web-based electronic-tendering system — while mentoring colleagues in MVVM, LINQ, and application-design patterns.
Education
M.Eng., Computer Science – Machine Learning
M.Sc., Software Engineering
M.Sc., IT Management
Publications
GuideArch: Guiding the Exploration of Architectural Solution Space under Uncertainty. With Naeem Esfahani and Sam Malek. ICSE, San Francisco, 2013.
Dealing with Uncertainty in Early Software Architecture. With Naeem Esfahani and Sam Malek. FSE-20, NIER track, Cary, NC, 2012.
Certifications
DeepLearning.AI
DSPy: Build and Optimize Agentic Apps!Jun 2025
LLMOpsJan 2025
Functions, Tools and Agents with LangChainJan 2025
Intro to Federated LearningDec 2024
Understanding and Applying Text EmbeddingsDec 2024
Reinforcement Learning From Human FeedbackDec 2024
How Diffusion Models WorkDec 2024
Building Applications with Vector DatabasesDec 2024
Knowledge Graphs for RAGDec 2024
Reasoning with o1Dec 2024
Generative AI: Working with Large Language ModelsSep 2023
Learning Amazon SageMakerSep 2023
PyTorch Essential Training: Deep LearningAug 2023
Machine Learning with Python: FoundationsAug 2023
Skills
Languages
Data engineering & storage
Machine learning
Generative AI
AI-assisted engineering
Cloud, DevOps & tooling
Backend, UI & architecture
Crossings
Every gardener wanders — and every garden receives guests. One world below carries both stories: in gold, my own route through it, gems for the cities I've lived in and points for the ones I've passed through; in green patina, the countries you — its visitors — arrive from, each shaded by how many. Give it a spin.