Maitar Asher is a founding engineer and Head of Engineering at Tavily, building infrastructure that gives AI agents access to current information across the web. Her work spans search retrieval, intelligent caching, and the evaluation systems needed to determine whether web-connected AI produces reliable, well-grounded answers.
Asher studied computer science at Columbia University and joined its School of General Studies Honor Society in 2023. At Columbia, she and future Tavily founder Rotem Weiss built an early research agent together; their project earned a B-minus. Her earlier experience included product and operations work at Lazzy, teaching artificial intelligence at Columbia, and machine-learning research engineering at Stanford.
At Tavily, Asher helped build real-time web search for AI agents serving applications such as legal research, sports information, and financial-fraud detection. Following Nebius’s acquisition of Tavily, that search infrastructure became part of a broader production AI platform.
- Dynamic evaluation datasets: Static benchmarks such as SimpleQA and HotpotQA cannot fully test answers dependent on changing information, source quality, timing, and user context. Asher advocates regularly refreshed, domain-specific question-and-answer datasets grounded in live web documents.
- Multi-provider grounding: Evaluation data assembled exclusively from one search provider can flatter that provider’s retrieval system. Asher favors gathering evidence across multiple providers, preserving source context, and tracking experiments through LangGraph and LangSmith. Tavily’s related open-source evaluation generator was initiated by its data leader, Eyal Ben Barouch.
- Holistic search evaluation: Search quality includes source diversity, document relevance, hallucination risk, and accuracy. Asher also supports reference-free evaluation when labeled answers are unavailable, while distinguishing her engineering perspective from the specific metrics developed by her fellow panelists.
