Cooma AI research & analysis

Insights

Field notes on the hardware, models and economics behind AI in production.

A strip-chart recorder traces a signal onto folded graph paper.

Articles

  1. Heterogeneous Inference: Building a More Efficient Token Factory

    How prefill/decode separation, hardware matching and cache-aware scheduling shape the economics of AI inference.

  1. Beyond GPU Rental: Turning AI Tokens into Business Outcomes

    Why enterprise AI buyers need more than cheaper compute—and how governed model access changes the decision.

  1. The Token Factory Flywheel: Making Every AI Workflow Better

    How repeatable workflows, reviewed exceptions and measured pilots turn model access into business value.

  1. When should an AI agent stop and ask?

    A purchase order can time out after the supplier has accepted it. See where to put approval, what a stop button can undo and how to measure the cost of supervision.

  1. What does Australian AI capacity mean for a buyer?

    A maintenance assistant puts an Australian hosting requirement to the test: follow one fault report through retrieval, model inference, support and a ten-minute GPU outage.

  1. A model fits on one GPU. How many users fit with it?

    OpenAI showed that gpt-oss-120b could run on one 80 GB GPU. A burst of twenty contract reviews shows what a buyer must test before promising a two-minute turnaround.

  1. Which sites should a field team inspect first?

    Use satellite embeddings to choose field visits, then compare travel, GPU time and model API costs. A worked example shows where each kind of compute earns its place.

  1. The citation is real. Does it support the answer?

    A contract assistant can quote the right clause and still give the wrong answer. A worked review queue shows when improving the evidence saves more than changing models.

  2. Which DeepSeek-R1 are you testing?

    An invoice-review trial compares DeepSeek-R1 checkpoints by correction cost, latency and routing errors. A model that costs more per call can cost less per resolved case.

  1. What should an AI assistant be allowed to look up?

    A warehouse stock lookup shows how to set an MCP tool’s permissions, trim its token bill and keep a model from promising inventory the caller cannot reserve.

  1. When the same prompt costs less the second time

    An October 2024 API bill, rebuilt line by line, shows when prompt caching cuts total spend and why the share saved falls as answers get longer.