Philip Kiely writes about inference engineering and works on special projects at Baseten. He is the author of Inference Engineering and Writing for Software Developers, connecting technical implementation with explanations developers can use.[1][2][3]
Start with Inference Engineering →
X 
GitHub Background
Kiely earned a computer-science BA with honors at Grinnell College in 2020. His first technical book grew out of writing tutorials and being asked how developers could publish similar work. His work at Baseten combines software, technical communication and projects that explain how generative models are served in production.[3]
Contributions
A map of the inference stack
Inference Engineering organizes model serving into runtime, infrastructure and tooling. It starts with product requirements and model evaluation, then connects model architecture and hardware constraints to serving software and optimization techniques. The book covers quantization, speculative decoding, key-value-cache reuse and distributed execution, before examining production operation. This gives readers a way to connect a performance problem with the layer of the system that can address it.[2]
Teach developers to structure useful writing
Writing for Software Developers treats technical writing as a craft developers can practice. It draws on Kiely’s experience publishing tutorials and explains how to turn a project or technical lesson into material another developer can follow. The emphasis connects technical knowledge with clear structure and the reader’s practical task.[3]
Ideas
In his 2026 update talk, Kiely contrasts fitting a model onto available local hardware with improving throughput and latency in a datacenter. Local deployment often starts with compression; larger serving systems must also coordinate requests, cache state and accelerators. He uses this distinction to organize newer techniques and to ask what has changed since the book, while retaining the underlying goal of reliable, efficient model execution.[4]
Tools, writing & talks
Writing, research & talks
Start here
Inference Engineeringbaseten.comA book connecting models, hardware, serving software and production optimization.
Start with the opening map and prerequisites chapter before choosing an optimization technique. Basic familiarity with deploying a model helps; the sequence connects a product’s latency and cost requirements to the serving stack.
Author[2]Baseten's Philip Kiely walks through what changed in inference engineering: 4-bit KV cache trades bandwidth for compute in ways that only pay off on memory-starved local hardware, and diffusion-based speculative decoding more than tripled acceptance in production.
Catalog workOriginal source- What’s New in Inference Engineeringai.engineer
A conference update on newer model-serving techniques and deployment tradeoffs.
Speaker[4] - Writing for Software Developerswfsd.com
A book about developing and publishing useful technical explanations.
Author[3]