Hardware & Systems· September 2026· 4 min read
GLM-5.3-Flash and Qwen3.8-Flash-Next: what fits on a desk
What running 320B and 180B-class multimodal open weights means for local DGX Spark and workstation hardware.
08 · WRITING
Writing
Deep-dive technical guides, memory-leak postmortems, and hardware benchmarks from our engineering lab.
Guide & Systems Architecture· August 2026· 12 min read
Running Open-Source AI: The Complete Hardware Guide from Home Lab to Production Workstation
How to run open-source AI locally at home or in the office. Real VRAM requirements, NVFP4 vs BF16 quantization, token throughput benchmarks, and choosing between consumer GPUs and on-premise workstations.
Field Notes & Systems· August 2026· 9 min read
I picked the wrong one twice
Three code-generation prompts, run against the same model on my own hardware and on the vendor's hosted service, both at maximum reasoning effort. I judged the results by looking at them. I got two of the three wrong, and both times the output I preferred was the one with a bug in it.
Hardware & Systems· August 2026· 7 min read
Running DeepSeek-V4-Flash-0731 on One DGX Spark With Full Quality
A fully local reasoning stack on a single GB10 box, the benchmark numbers, and the memory-leak hunt it took to get there.
Legal AI· August 2026· 4 min read
Why legal teams are moving AI closer to their documents
How local retrieval lets firms work across contracts, discovery, and matter files without sending privileged material into external AI systems.
Hardware & Systems· August 2026· 6 min read
One Spark, dual Spark, or custom RTX PRO: which path, when
When one Spark is enough, when linking two enables frontier models, and when a custom RTX PRO workstation or multi-GPU server is the honest recommendation.
RAG & Retrieval· August 2026· 5 min read
RAG is the first local workload most teams should test
Why document search, summarization, and source-grounded Q&A are the cleanest starting points for private AI in regulated industries.