https://www.ovidiudan.com/Ovi's Tech BlogThese are my experiments with AI, LLMs, and shipping useful services. 2026-10-04T10:52:26+00:00 Ovidiu Dan https://www.ovidiudan.com/ Jekyll © 2026 Ovidiu Dan /assets/img/favicons/favicon.ico /assets/img/favicons/favicon-96x96.png Unreasonable things I ask AI to do2026-10-04T10:35:00+00:00 2026-10-04T10:52:06+00:00 https://www.ovidiudan.com/2026/10/04/unreasonable-things-i-ask-ai-to-do.html Ovidiu Dan Getting a museum audio tour out of a region-locked Android app, mapping Xfinity upgrades around my house, and looking for a fiber cable at nearby stores. A few things I have been asking AI agents to do from my phone. Planning a four-GPU expansion box for my AM5 workstation2026-09-17T05:43:00+00:00 2026-09-17T05:43:59+00:00 https://www.ovidiudan.com/2026/09/17/gpu-expansion-box-pcie-switch.html Ovidiu Dan My plan to connect an AM5 workstation to four GPUs through a Broadcom PEX88096 backplane, with PCIe P2P between the cards and a shared x16 connection to the CPU. Combining GPT-6 Astra intelligence with local AI grunt work2026-09-13T18:30:00+00:00 2026-09-13T18:30:00+00:00 https://www.ovidiudan.com/2026/09/13/astra-orchestrator-local-subagents.html Ovidiu Dan One week of GPT-6 Astra directing local pi subagents through CUDA kernel research on a $100/month OpenAI Pro plan: $475.97 of metered-equivalent orchestrator usage that saturated my weekly limit, plus $1,416.57 of work done locally that the plan had no room for. The enabler is a 2 KiB RETURN.json cap that kept 29 GB of worker transcript out of the orchestrator's context. My personal LLM toolchain: switch between models and harnesses, control them from your phone2026-08-27T20:30:00+00:00 2026-08-27T20:57:59+00:00 https://www.ovidiudan.com/2026/08/27/my-personal-llm-toolchain.html Ovidiu Dan Paseo to run and watch agents from any device, Pi as the harness, and four extensions (web search via Gemini, Chrome, sudo, sub-agents) that make local models useful. Recipe for running Qwen3.8-Flash-Next in NVFP4 on 2x RTX Pro 60002026-08-27T15:00:00+00:00 2026-08-27T15:51:03+00:00 https://www.ovidiudan.com/2026/08/27/recipe-qwen38-flash-next-nvfp4-rtx-pro-6000.html Ovidiu Dan Qwen3.8-Flash-Next narrates its own NVFP4 install on a 2x RTX Pro 6000 (Blackwell) workstation: the dedicated vLLM recipe image, tensor-parallel launch on sm_120, and the n-gram embedding table offloaded to system RAM so the weights finally fit in 192 GB of VRAM.