Enterprise Knowledge Search
Suntime needed an authoritative, source-traceable research assistant for professional investors, but its self-built retrieval stack struggled with the semantic complexity, latency, and operations burden created by huge volumes of financial reports, announcements, research notes, news, and fund data.
Reported impact
Suntime improved retrieval speed by more than 10x, cut operations work by 80%, and kept the service stable through more than 100 peak-traffic events over six months, giving researchers faster access to authoritative evidence while reducing the burden of maintaining a custom vector search stack.
Suntime built its AI Xiaoer investment-research assistant on top of Alibaba Cloud's Vector Search Service Milvus Edition to turn large financial document collections into a low-latency semantic retrieval layer. The architecture is organized in layers: raw announcements, research reports, financial statements, news, and fund data are collected at the bottom; parsing and embedding steps convert those documents into vectors; then Milvus stores and filters the vectors so the system can retrieve relevant evidence for natural-language questions. AI Xiaoer sits above that retrieval layer and uses the returned passages to power functions such as intelligent Q&A, rapid research, deep analysis, thematic recommendation, and report generation. The platform gives Suntime a scalable path from heterogeneous data ingestion to semantic retrieval and cited answer generation without the performance and maintenance limits of its prior PostgreSQL-based vector stack.
Financial announcements, reports, news, and fund data are parsed and embedded into vectors, stored in Milvus for semantic retrieval, and then retrieved passages are passed to AI Xiaoer so the assistant can answer investor questions, summarize evidence, and generate research outputs with traceable source material.
AI Xiaoer combines generative AI with Alibaba Cloud Milvus to deliver intelligent Q&A, rapid research, deep analysis, thematic recommendations, and report generation (original: AI 小二基于生成式 AI 能力，结合阿里云向量检索服务 Milvus 版... 打造了集“智能问答、极速研究、深度分析、主题推荐、报表生成”于一体的智能化投研平台).
Across more than one hundred peak-traffic events over half a year, Alibaba Cloud Milvus maintained 0 failures and 0 interruptions (original: 阿里云 Milvus 历经半年百余次瞬时访问峰值考验，始终保持 0 故障、0 中断).
Average response time fell from 600ms on the open-source PostgreSQL vector stack to 50ms on Alibaba Cloud Milvus (original: 开源 PG 库平均响应速度 600ms，阿里云 Milvus 平均响应速度 50ms).
Suntime had to manage unstructured and semi-structured financial data such as announcements, reports, news, and fund data, and traditional keyword search could not understand semantic relationships well enough (original: 朝阳永续需处理包括上市公司公告、财报、卖方研究报告、新闻资讯、基金数据在内的海量非结构化与半结构化数据... 传统的关键词检索方式难以理解语义关联).
The system uses a layered architecture in which parsed and embedded documents are stored in Milvus and exposed through standardized low-latency retrieval APIs (original: 底层... 中间层... 并存入阿里云 Milvus；上层... 实现多模态召回；接入层：通过标准化接口为 AI 小二提供低延迟、高并发的检索服务).
Review how this reference was scored and trace each normalized claim back to its documented source.
Scoring methodology
Customer Reference Value measures usefulness as a customer reference. Evidence Confidence measures the quality and traceability of available evidence, not independent verification.
Traceability
Field coverage shows which facts each source supports. Dates are shown only when documented by the source.
The introduction page identifies the company as 上海朝阳永续信息技术股份有限公司 and says it is 中国领先的金融数据服务商, supporting a headquarters/operating region in China.
Related
MongoDB Sample Oct 8, 2026 Navisio AI Rapid adoption of Navisio AI's platform triggered severe latency spikes, elevated CPU and RAM consumption, and costly vertical scaling. As it prepares to scale to 1,000 by the end of the year, and eventually toward 5,000 users, the team is… MongoDB Atlas · MongoDB Atlas Vector Search 
Google Cloud Sample Archived · Sep 27, 2026 SIGNAL IDUNA SIGNAL IDUNA needed to spread AI safely across a large insurance workforce, make a century of internal policy knowledge searchable, and let teams build their own agents without creating a fragmented, insecure shadow-AI… 80% weekly active adoption rate with a +45 NPS in first five months Document AI · Gemini · Gemini Enterprise Agent Platform Google Cloud Sample Archived · Sep 25, 2026 Gain However, as Brian Leenen points out,"The challenge goes beyond automated data collection. By combining automation with human validation, Gain has successfully scaled its news production. Its pipeline… AlloyDB · AlloyDB for PostgreSQL · Cloud Run 
