Build and operate first-class agent modalities on the platform, including voice agents (speech-to-text, text-to-speech, low-latency streaming, and turn-taking), document-extraction agents (parsing, OCR, structured field and table extraction from complex documents), and agentic memory (short- and long-term memory, persistence, retrieval, and context management across sessions). Drive systematic hill-climbing of core agent capabilities-including NL2SQL, RAG, voice, document extraction, and tool use-by building evaluation datasets, benchmarks, and quality metrics, then iterating on prompts, retrieval, and orchestration to measurably improve accuracy.