Content

AI, XR, DevOps, graph theory, knowledge graphs, and digital sovereignty. Deep dives, tutorials, and analysis from across the GraphWiz knowledge base.

Benchmarking pi-l1-cache: a 138µs in-memory cache layer for the pi coding agent

We benchmark pi-l1-cache, a zero-dependency L1 in-memory cache extension for the pi coding agent. 138µs per request, honest about where the 100x hash claim breaks.

pillm-cachingperformanceextensions+1 more

August 19, 2026· ~11 min read

DeepSeek V4 Flash 0731: Initial Experiences on 2x DGX Spark

Practical results from serving DeepSeek's latest MoE mixture-of-experts model on dual DGX Spark systems with vLLM, DSpark speculative decoding, and Anemll optimizations.

deepseekv4-flash

moe

dgx-spark

+5 more

August 18, 2026· ~8 min read

Rust GPU Offload: Portable, Safe, and Fast - Paper Analysis

Deep analysis of arXiv 2608.13759: Rust Offload brings CUDA/HIP-level GPU programming to Rust without vendor lock-in, using LLVM Offload for cross-vendor portability and Rust's type system for memory safety in kernels.

rustgpucudahip+5 more

August 17, 2026· ~4 min read

GRALAN: When Knowledge Graphs Speak LLM Native

GRALAN enables KGs to speak directly in the LLM's semantic space through relational tokens that preserve graph structure. No fine-tuning required — just trainable language mediation.

knowledge-graphsllmgraph-language-modelsrelational-tokens+1 more

August 17, 2026· ~3 min read

X-KGRank: When Small Language Models Lie (And How Knowledge Graphs Fix It)

X-KGRank unifies structural collaborative filtering with LLM-based explanation. A 1.5B parameter model matches a 7B model on explanation quality — but fabricates facts more often when not grounded.

recommendation-systemsknowledge-graphsexplainable-airag+1 more

August 17, 2026· ~5 min read

DyCoder: How Human Developers Think About Code Dependencies

Human developers build partial dependency graphs on-demand — not static global graphs. DyRetriever replicates this pattern, achieving +59.73% Pass@1 on DevEval while being 7.4x faster.

code-generationdependency-graphsragllm+1 more