Fiodar Sazanavets on Production-Grade Python, LangChain & LangGraphIn this episode of the Leanpub Podcast, host Len Epp speaks with Fiodar Sazanavets, a senior software engineer, former Microsoft engineer, and four-time Microsoft MVP, about his Leanpub book Production-Grade Python, LangChain & LangGraph.From Belarus to British Software EngineerFiodar grew up in Belarus and moved to the UK in 2001 due to political circumstances. He describes the economic chaos of post-Soviet Belarus, including hyperinflation on the scale of Zimbabwe, and how arriving in the UK gave him a sharpened sensitivity to opportunity that many around him seemed to lack. His academic background was unconventional for a software engineer: a bachelor’s degree in biology and a master’s in environmental informatics, which led to his first job as a flood risk analyst. Finding that career a dead end financially, he taught himself to code starting with VBA in Excel, then VB.NET, eventually building a decade-long career in the .NET and C# ecosystem.Building a Career as an Educator and AuthorFiodar began teaching on Udemy in 2018 and was subsequently approached by technical publisher Packt to narrate code samples from books. He also wrote long-form articles for John Sonmez’s Simple Programmer platform. In 2021, he wrote two books simultaneously — a mindset book called Battle Hardened Developer for Simple Programmer and a technical book for Packt. With the manuscripts in their publishers’ release pipelines, he discovered Leanpub through an article by bestselling Leanpub author Jeff Geerling, who wrote about his experience self-publishing a book on Ansible. Inspired, Fiodar quickly converted an existing Udemy course into his first Leanpub book, SignalR on .NET 6: The Complete Guide, and found it more successful than expected. He has since published additional titles, including The Easiest Way to Learn Design Patterns, and praises Leanpub’s Markdown-based writing workflow, WYSIWYG editor, and print-ready PDF export feature.Why This Book ExistsAs AI engineering became the dominant force reshaping software development, Fiodar noticed that Python’s ecosystem for AI was far more mature than .NET’s. Running his own consultancy, Orion AI Engineering, he wanted to serve clients with the best available tools rather than being constrained by his own specialisation. He began studying production-grade Python intensively — not the syntax, which he already knew, but the engineering fundamentals: project structure, dependency management, FastAPI, async programming, testing, deployment, and observability. He accumulated extensive notes and realised that many developers with basic Python knowledge faced the exact same gap. The book is, in his words, “the book I wish I had when I started.”Production Engineering in the Age of AIA central theme of the conversation is why deep engineering knowledge remains valuable even as AI tools become capable of generating large amounts of code. Fiodar argues that AI models are trained on average code — mostly hobby projects and tutorial examples from public repositories — and will produce average results unless guided by someone who understands production requirements. Non-functional requirements such as scalability, performance, fault tolerance, and partial failure handling are invisible to those without engineering training, but are precisely what separates a working prototype from a reliable production system. He also emphasises that engineers remain accountable for AI-generated code and must maintain deterministic guardrails such as linters, static analysis tools like SonarQube, and structured code review practices.A Real-World Agentic SystemTo illustrate what readers can build, Fiodar describes an agentic system he built for a company with a large industrial data platform. An orchestrator agent routes user queries to specialised sub-agents — one for retrieving documents, another for time-series data such as wind speed measurements — with a final consolidation agent that combines responses and applies hallucination prevention before returning output to the user. LangGraph handles orchestration and session checkpointing, while LangChain manages direct LLM interaction and memory. This architecture, he explains, works better than a single agent because language models handle narrow contexts more reliably than broad, multi-concept ones.Writing Process and Leanpub WorkflowFiodar describes his writing process as analogous to software development: outlining first, then building the book chapter by chapter, treating it as a product to be engineered rather than a literary endeavour. His current book has around 60 short chapters averaging three pages each, keeping concepts digestible. He prepares a companion repository before writing, in this case a realistic production-grade example of a Python project with LangChain and LangGraph implementing an agentic IT support ticket monitoring system. He uses AI tools for proofreading and code review and applies the same deterministic quality tools he recommends in the book. When asked for a feature request, he nominated integrated print-on-demand publishing, a wish shared by many Leanpub authors.This interview was recorded on [interview-date].The full audio for the interview is here: https://s3.amazonaws.com/leanpub_podcasts/video1091098144-edited.mp3. The Leanpub Podcast is available on our YouTube channel at https://www.youtube.com/leanpub, in Apple Podcasts here https://podcasts.apple.com/ca/podcast/Leanpub/id517117137, and almost everywhere else people listen to podcasts.