Thoughts on AI infrastructure, CTV development, and production engineering.
How we built an AI agent framework with intent routing, MCP integration, and a 600K+ node knowledge graph - including lessons learned from deploying on Intel Arc GPUs.
Challenges and solutions for maintaining feature parity across Roku, FireTV, Android TV, Tizen, webOS, and Vizio - from shared components to platform-specific CI/CD.
Deploying vLLM with tensor parallelism across 4 Intel Arc GPUs - prefill benchmarks, KV cache management, NVFP4 quantization results, and the --entrypoint /bin/bash gotcha.