While traditional search engines are still stuck on keyword matching, @SentientAGI's core researcher Salah Alzu'bi is already redefining the complex challenges of AI system thinking. From earning a Master's in Computer Science at UMass to contributing to Google DeepMind, Meta FAIR, and Microsoft Research, Salah's research journey has always revolved around enabling AI models to truly understand and tackle long-term tasks. ⚡️ ROMA: The Recursive Revolution of Multi-Agent Systems ————————————————————————— A single AI handling complex tasks is like one person trying to do multiple things at once. The ROMA framework, led by Salah, has completely transformed this scenario. Recursive decomposition control: complex queries → subtasks → atomic execution → result aggregation This isn’t just simple task allocation but intelligent hierarchical coordination. Each node can autonomously decide whether to execute directly or further decompose, with the entire process being transparent and traceable. In the SEALQA Seal-0 benchmark test, ROMA Search achieved 45.6% accuracy, surpassing the previous best Kimi Researcher (36%) and more than doubling Gemini 2.5 Pro (19.8%). Long-Term Tasks ————————————————————————— Salah’s core focus is addressing AI’s short-sightedness. Traditional models excel at single-step reasoning but fail when faced with complex tasks requiring multiple steps, tools, and time. ROMA’s recursive control mechanism equips AI with project management capabilities: breaking down goals, allocating resources, coordinating execution, and integrating results. This fundamentally changes cognitive architecture. When AI systems can collaborate, think recursively, and divide tasks like human teams, true long-term autonomous task execution becomes possible. @SentientAGI is redefining the boundaries of AI systems from three dimensions: evaluation, coordination, and search. As multi-agent systems gain reliable evaluation, recursive coordination, and deep search capabilities, we’re one step closer to achieving true AGI.