This repository is a hybrid research-operations codebase for the AMD BC-250 (Cyan Skillfish APU), a repurposed crypto-mining board with 16GB UMA. The dominant character is an application: a collection of hundreds of Python and shell scripts for LLM inference benchmarking, personal home-lab automation, OSINT monitoring, and interactive Signal bot functionality. The system runs on a single machine (bc250) with Ollama, llama.cpp, stable-diffusion.cpp, and signal-cli as backends. Benchmark scripts measure generation/prefill throughput, context scaling, KV-cache quality, and hardware roofline, producing extensive JSON results that feed a research paper pipeline (LaTeX tables, provenance auditing). The netscan directory implements a personal intelligence platform: network security scanning, career/patent/news scraping with LLM analysis, leak monitoring via dark web sources, and Home Assistant integration. While the code is tightly coupled to the specific hardware, model versions, and the author's personal environment (hardcoded paths, phone numbers, systemd services), it demonstrates deep systems engineering knowledge and a systematic approach to benchmarking. Reuse outside this environment would require significant adaptation, but the benchmark methodologies and some isolated scripts (e.g., hardware microbenchmark, GGUF parser, SinoTrack GPS tracker reverse-engineering) could be extracted with moderate effort.