Fanout field notes
Notes on AI research, systems, inference, and machine learning mathematics.
Follow via RSSPublished notes
Recursive Language Models
Treat long prompts as a REPL variable and recursively query the model over snippets — scaling past context windows without stuffing the transformer.
Kimi K3 architecture: KDA, MLA, and MoE explained
A walk through a modern frontier model's architecture: linear attention variants, latent attention, and sparse expert routing.
The KV cache formula: LLM inference memory
Deriving the memory cost of attention state and why it dominates long-context serving.
How AI memory works: five systems
From context windows to retrieval indexes: the distinct mechanisms people call 'memory'.
Machine learning math for software engineers
The minimum rigorous math toolkit, ordered by what an engineer actually uses first.
Deep learning roadmap for self-taught engineers
A sequenced path from first principles to reading current papers.
System design interview prep for backend engineers
Turning production experience into structured architecture answers.
ML engineer interviews as a software engineer
What ML loops test that SWE loops don't, and how to close the gap.
A research paper reading habit for staff engineers
Fifteen papers a quarter, deliberately chosen, beats one hundred skimmed abstracts.
AI research papers for system design engineers
The papers that matter most when your background is distributed systems.
Daily paper summaries for AI research engineers
How the Daily Fanout format compresses a paper into a five-minute read.
Deep learning for computer vision engineers
Bridging classical CV pipelines into learned representations.
ML engineer roadmap for career switchers
Sequencing the switch: math, modeling, and production skills in dependency order.
Machine learning roadmap for backend engineers
Reusing your engineering strengths while filling modeling gaps.
ML interview prep for data scientists
From analysis to engineering: what to add before interviewing for MLE roles.
ML math roadmap without a math degree
Sets, linear algebra, calculus, and probability in the order ML actually consumes them.
System design interview prep for new grads
Building architecture intuition without years of production scars.
System design labs for distributed systems engineers
Small executable builds that make distributed-systems concepts concrete.
System design prep for senior engineer interviews
What changes when the loop expects judgment, not definitions.