Stephen Shary
Principal Engineer at ACV Auctions
Columbus, Ohio, United States
Actions
I am a full-stack developer and technical lead that uses Python, JVM languages and Javacript. I enjoy functional programming, delivering production code as fast as possible and robust automated testing. I work with teams to overhaul their projects, improve code quality, speed delivery to production and improve culture.
I am also working in machine learning and AI to improve areas ability to provide accurate data and supplement gaps with predictions. I focus on neural networks with Keras and Tensorflow. I am learning how to productionalize data science to become a platform for companies and departments
Links
Area of Expertise
Topics
Exotic Agentic Development Strategies: Everything old is new again
For 40 years we knew how to write software correctly. We just couldn't afford to. Now we can.. We all know you can vibe out some slop on a grammatically incorrect, partial coherent, stream of consciousness. The question is: how do we use AI to develop code at a higher quality level? How do we utilize Ai to write code better than we could. Not more, but better. With the advent of cheaper coding, we should take a look back at previous failed methods of doing software engineering when coding was difficult. Many past practices did not become popular because the cost to benefit was much too high. With that balance changing, we have a treasure trove of ideas and techniques to improve agentic development.
In this talk, explore multiple techniques and how to apply them to a agent/harness. We will investigate design, review and verification ideas that have the proven ability to reduce bugs by 90%, improve reliability to a mathematical certainty and implement not only a coherent design, but one that meets all non-functional constraints. Techniques used by JPL and NASA for maximal system reliability. Ideas promoted by Knuth and Meyers in the 1980's and 1990's that are fantastic from a theoretical perspective. Model driven development, property based testing, and other techniques that are now trivial with AI.
Everything old is new again, but now affordable.
Architecting for AI: Building Production-Ready Systems with Context, Control, and Automated Testing
The explosion of AI capabilities, particularly with Large Language Models (LLMs), has created both incredible opportunities and significant architectural challenges for experienced software professionals. While conversational interfaces capture the headlines, realizing the true potential of AI requires building programmatic systems that can automate complex tasks and drive business value. This talk dives deep into the architectural patterns and techniques needed to move beyond the chatbot demo and build reliable, production-ready AI applications, specifically addressing the pitfalls of general knowledge LLMs.
The key to unlocking LLMs' true potential lies in supplementing their core knowledge with specialized, context-specific information. We’ll explore strategies for enriching the LLM’s understanding before it engages with a prompt – a crucial step in ensuring accurate and reliable responses. This leads us to examine a range of knowledge integration architectures, including RAG, CAG, and KGR. We'll also look at the architecture of Chain-of-Thought (CoT) and how that can improve deep thought problems. We will also look at architecting different domains of knowledge and function through agentic AI.
Critically, we’ll examine architectures designed to limit both input and output, enabling early detection and mitigation of hallucinations – a key concern for production AI systems. You'll learn how to enforce boundaries and control LLM behavior to ensure predictable and trustworthy results.
Finally, we'll introduce a powerful technique: using AI to generate and validate test use cases, demonstrating a Monte Carlo sampling process for robust testing. This allows for efficient monitoring of progress and rapid identification of regressions, ensuring high-quality AI-powered solutions. This talk is designed for software architects, developers, and technical leaders seeking to effectively integrate LLMs into their applications. Attendees will gain a practical understanding of how to build scalable, testable, and reliable AI-powered solutions, ready to tackle real-world challenges. You'll leave with actionable strategies focused on delivering AI solutions with verifiable accuracy and controllability.
Master All Frontends in Real-Time with React Native
Front-end development is fragmented with many different stacks that are tailored to: web, Android, or iOS. We demonstrate live how React Native can be used to develop on all three platforms at the same time. We will show how to kick start a new project, make it run on all three platforms and then also (with some luck) deploy it on the phones in the audience. Learn the pros and cons of using a cross-platform technology to solve feature parity, infrequent deployments, and disparate developer skillsets.
Advancing your Technical Career Beyond Sr. Engineer
At times, it feels like everyone qualifies to be a Sr. Engineer. After a year of experience, many companies can't really distinguish between engineering levels. It can be difficult for engineers to understand how to grow in a technical career. Years of experience are the only real factor. In this talk, we present technical paths beyond a senior engineering level. We present companies that promote different growth tracks along with engineers that distinguish themselves. This will provide actionable opportunities along with ideas on how to identify when your current role won't grow and how to respond.
Continuous Performance Testing
Performance testing can be one of the most complicated parts of testing to execute and evaluate. In this talk, learn about ways to build it into your pipeline. Understand how to properly tune and assert on test results. We will use locust.io to present the usage of performance testing in our pipelines at ACV and how it has helped our devops culture.
Patterns and Architectures Beyond Microservices
Have you slayed the monolith and are ready for more? Learn patterns and architectures that move beyond simple microservices to solve problems of resiliency, security, scalability, responsiveness. Understand the new class of issues that come with distributed systems like microservices. We explore modern patterns like CQRS, sidecars, probabilistic development, code as documentation and others. You will be able to see past the monolith and understand the new problems and patterns that help in modern distributed systems.
Level Up Your Network Infrastructure with Automated Testing
We have arrived at predictable infrastructure. At Kroger, we use Nginx as a reverse proxy to route traffic with a configuration that is thousands of lines long with pull requests from dozens of teams. Even with this scale, we are still able to complete on-demand deployments to multiple environments that serve dozens of domains and route to over 50 applications clusters with zero downtime. We show an open source framework, SnowGlobe, which allows us to simulate and test every possible traffic routing situation before we deploy. We use a full CI/CD pipeline with 7,000+ tests run for every commit.
Stir Trek 2025 Sessionize Event
Momentum 2024 Sessionize Event
Momentum 2023 Sessionize Event
CodeMash 2023 Sessionize Event
Momentum 2022 Sessionize Event
DevOps Days Buffalo Sessionize Event
CodeMash 2022 Sessionize Event
Momentum 2021 Sessionize Event
CodeMash 2019 Sessionize Event
Stir Trek 2018 Sessionize Event
Stephen Shary
Principal Engineer at ACV Auctions
Columbus, Ohio, United States
Links
Actions
Please note that Sessionize is not responsible for the accuracy or validity of the data provided by speakers. If you suspect this profile to be fake or spam, please let us know.
Jump to top