Hardware-first observability and cost governance for self-hosted inference — the MPG rating for local AI. Tracks WES (a tok/W efficiency score with thermal honesty) and attributes real $/1M-token cost across your own GPUs, so “local is cheaper” is a number you can show, not an assumption. Community tier free.
Personal research, not a product. Open methodology for building private
enterprise language models: the thesis is that a 7B model trained on the
right data beats a frontier model on a narrow task and runs locally for
free. Current work: account-intelligence-7b-v1 — fine-tuned Qwen2.5-7B for
enterprise account briefs across six surfaces.
Agentic enterprise platform running Qwen2.5-32B-FP8 on a DGX Spark alongside Claude. Generates account intelligence, meeting prep, and pricing assistance for sales teams. Peers into Salesforce, Confluence, and internal systems in real time.