LLM-Powered Data Pipelines

In this episode of TechUnhinged, host Rabia Javeed sits down with Chris Colinsky, CTO and Principal AI Engineer at Howl, which has been building for the web since 2003 and is the creator of the open-source OpenArmature framework for LLM-powered pipelines. Chris cuts through the AI hype, arguing that large language models are pattern matching at scale, probabilistic, confidently wrong, and only as reliable as the engineering around them. He explains why most AI failures are really data quality problems, why legacy system modernization must precede AI adoption, and how his "calculate first, reason second" approach builds reliable AI systems on unpredictable models. From graph-based retrieval to self-hosted small models, this conversation is a practical playbook for taking AI from demo to production.