Llm

At Google I/O Extended Newport Beach, I watched Gemma 4 go from a 147-second response down to a few seconds — with no new hardware, just five inference optimization techniques stacked on top of each other. Here is what each one actually does, in plain language.

I benchmarked Claude Sonnet 4.6, Gemini 2.5 Pro, and Qwen3 Max Thinking on translating bilingual kitchen transcripts into valid culinary JSON schemas. Here is how they handle complex schemas, Konglish loanwords, and cultural nuances.

Honored to participate in the AI Valley Hackathon in Palo Alto! Our team developed Oppalingo, an AI-powered K-drama-inspired language learning app, which made it to the finals by using Whisper API to provide custom feedback on users' audio input.