SBIR Phase II: Pushing the Boundaries of Intelligent Assistants for Financial Services

小型企业创新研究(SBIR)第二阶段项目的更广泛影响/商业潜力在于提供最先进的工具，使任何人都可以构建和部署特定于领域的商业智能虚拟助理(IVA)解决方案。这些工具允许其他人了解IVA应该如何构建，并将IVA技术集成到他们的产品中。IVA已在金融服务、医疗保健、教育、执法和零售等众多商业领域显示出希望，通过为人们提供与复杂的计算机和信息系统进行自然对话的媒介，减少了领域内获取知识的障碍。这个小型企业创新研究(SBIR)第二阶段项目将解决扩展智能虚拟助理(IVA)的域和功能时涉及的重大技术挑战。该项目将在设计能够学习和识别数百或数千个已学习概念的可扩展人工智能模型方面进行创新，并设计相应的系统架构，以支持日益增长的复杂算法的计算需求。具体地说，该项目旨在实现：(1)可扩展智能：处理成百上千种能力和可提取的语义概念的能力，允许用户用不受约束的语言与系统交互；(2)可定制智能：允许客户根据需要(半)自动培训和(重新)培训和定制智能的能力(添加新能力、识别新的槽值对、修改响应)；(3)会话复杂性：支持多轮对话，其中使用来自先前话语的上下文来改进和理解最终用户试图实现的目标；和(4)可扩展的系统基础设施：增强开源IVA软件基础设施，以根据负载无缝地扩展和缩小分配给每个智能引擎的计算资源。

The broader impact/commercial potential of this Small Business Innovation Research (SBIR) Phase II project is in providing state-of-the-art tools allowing anyone to build and deploy domain specific commercial intelligent virtual assistant (IVA) solutions. These tools allow others to understand how IVAs should be architected and integrate IVA technology into their offerings. IVAs have shown promise in numerous commercial domains including financial services, healthcare, education, law enforcement, and retail, to name a few, reducing the barrier to knowledge access within domains by providing a medium for people to converse naturally with sophisticated computer and information systems.This Small Business Innovation Research (SBIR) Phase II project will address the significant technological challenges involved when scaling the domains and capabilities of an Intelligent Virtual Assistant (IVA). This project will innovate in designing scalable artificial intelligence models capable of learning and identifying hundreds or thousands of learned concepts, and designing the accompanying system architecture to support the growing compute demand of sophisticated algorithms. Specifically, the project aims at achieving: (1) Scalable Intelligence: the ability to handle hundreds or thousands of competencies and extractable semantic concepts, allowing users to interact with the system with unbounded, unconstrained language; (2) Customizable Intelligence: the ability to allow customers to (semi-) automatically train and (re)train and customize the intelligence on demand (adding new competencies, identifying new slot-value pairs, modify responses); (3) Conversational Complexity: support multi-turn conversations, where the context from prior utterances is used to refine and understand what the end-user is trying to accomplish; and (4) Scalable System Infrastructure: enhance open source IVA software infrastructure to seamlessly scale up and down the computational resources allocated for each intelligence engine based on load.