SOBRE ESTE EPISÓDIO
Brandon Mathis sits down with Emerson Huitt to unpack when and why teams should run AI inference locally. They explore the business case for more control over sensitive data, privacy, compliance, and costs; compare on-premises hardware, developer machines, and rented GPU capacity; and share practical advice for testing workloads before investing in expensive infrastructure. The conversation also covers model selection, quantization, VRAM and cache planning, and evaluating local coding models for real-world development workflows.5 Key Points from this Episode:
- Local AI gives organizations more control over sensitive data, especially in regulated industries such as healthcare, defense, and businesses handling proprietary IP.
- The key value of local inference is not always lower cost, but more predictable cost, stronger privacy controls, and the ability to limit data leaving an approved environment.
- Start small before investing in hardware: rent GPU capacity, test real workloads, and determine whether on-premises, cloud-hosted, or developer-device inference fits the team’s needs.
- Model choice and quantization involve tradeoffs. Teams need to account for VRAM, context windows, performance, and quality, then test models against evaluations that reflect their actual workflows.
- Running local LLMs at team scale is an operations challenge. Reproducible environments, pinned versions, model checksums, secure sandboxes, and credential-injecting proxies are essential for reliability and security.
Brandon Mathis on Linkedin: https://www.linkedin.com/in/mathisbrandon/
Emerson Huitt on Linkedin: https://www.linkedin.com/in/emersonhuitt/
This Dot Labs Twitter: https://x.com/ThisDotLabs
This Dot Media Twitter: https://x.com/ThisDotMedia
This Dot Labs Instagram: https://www.instagram.com/thisdotlabs/
This Dot Labs Facebook: https://www.facebook.com/thisdot/
Sponsored by This Dot: https://ai.thisdot.co/