We believe agents will become autonomous in 12–18 months. We are building the research, protocols, and open-source tools they will rely on. 我们相信 Agent 将在 12–18 个月内实现自主化。我们正在构建它们所依赖的研究、协议和开源工具。
Peer‑review ready technical reports and open algorithms that power the AnswerMe platform. 在 AnswerMe 平台真实运行的核心算法与技术报告，面向同行评审和产业落地。
Large Language Model inference is a major computational and economic bottleneck. This paper presents a three‑tier semantic distributed caching architecture for LLM question‑answering systems: (1) a local user‑level LRU cache for personalized patterns, (2) a distributed Redis cluster with semantic hash indexing for cross‑user deduplication, and (3) a persistent knowledge graph store with vector‑based nearest‑neighbor search as semantic fallback. The AM‑MQM multi‑vector question matching algorithm is integrated to decide cache eligibility with 96% semantic accuracy. Deployed on the AnswerMe platform with 1.2 million monthly queries, the system achieves a 78% cache hit rate, reduces API costs by 91%, and cuts average latency from 2.3 seconds to 180 milliseconds for cache hits. 大模型推理的算力与成本压力日益增大。本文提出一个用于 LLM 问答系统的三层语义分布式缓存架构： (1) 用户级 LRU 本地缓存，捕获个性化模式； (2) 带语义哈希索引的分布式 Redis 集群，实现跨用户去重； (3) 结合向量近邻检索的持久化知识图谱存储，作为语义兜底。 通过集成 AM‑MQM 多向量问题匹配算法，在 AnswerMe 线上每月 120 万查询场景下， 缓存命中率达到 78%，API 成本降低 91%，命中请求平均延迟从 2.3 秒降至 180 毫秒。
AnswerRank is a twelve‑dimensional dynamic credit scoring framework for evaluating AI‑generated content credibility at scale. It integrates four layers: Call Quality (40%), Trust Propagation (30%), Temporal Value (20%), and Network Effect (10%), including a novel Enterprise Endorsement Index (EEI) and a four‑layer anti‑gaming mechanism. On a dataset of 5,000 AI‑generated answers, AnswerRank achieves 87% NDCG@10 in ranking quality, outperforming simple voting (62%), time‑weighted voting (71%), and PageRank variants (79%). Our anti‑gaming evaluation shows that manipulated answers experience a 65% score reduction while legitimate answers remain stable within 5% variance. AnswerRank 是一个面向大规模 AI 内容的十二维动态信用评分框架， 由四个层级构成：调用质量（40%）、信任传递（30%）、时间价值（20%）和网络效应（10%）。 框架提出企业背书指数（EEI），并设计四层反作弊机制。 在包含 5,000 条 AI 生成答案的数据集上，AnswerRank 的 NDCG@10 达到 87%， 显著优于简单投票（62%）、时间加权投票（71%）和 PageRank 变体（79%）。 反作弊实验表明，被操纵答案平均降分 65%，正常答案的波动控制在 5% 以内。
AM‑MQM (AnswerMe Multi‑Vector Question Matching) tackles question deduplication in AI‑powered knowledge systems by fusing four signals: semantic embeddings (45%), intent classification (20%), knowledge graph distance (20%), and scope normalization (15%). A three‑bucket scheme (SAME, FAMILY, DIFFERENT) enables graduated cache utilization strategies. Experiments demonstrate that AM‑MQM achieves 89.3% F1‑score on equivalence detection, outperforming Sentence‑BERT (82.1%) and SimCSE (84.7%). Our deployed system achieves a 78% cache hit rate while maintaining 96% semantic accuracy, reducing API costs by 73%. AM‑MQM（AnswerMe Multi‑Vector Question Matching）面向 AI 知识系统中的问题去重场景， 综合四类信号：语义向量相似度（45%）、意图分类（20%）、知识图谱距离（20%）和范围归一化（15%）。 算法提出 SAME / FAMILY / DIFFERENT 三桶策略，用于分级利用缓存。 实验表明，在问题等价检测任务上 F1‑score 达到 89.3%， 优于 Sentence‑BERT（82.1%）与 SimCSE（84.7%）。 在线部署后，在保持 96% 语义准确率的同时，缓存命中率达到 78%，API 成本降低 73%。
Six strategic products for the agent era, covering knowledge, creativity, and infrastructure. 面向 Agent 时代的六款战略产品，覆盖知识、创意与基础设施。
AI Answer Memory & Encyclopedia. Save, organize, and reuse AI conversations. Build your personal knowledge base. AI 答案记忆层与百科。存答案、用答案、AI 百科 —— 构建个人知识基础设施。
LiveAI-powered ticketing for arts and events. Smart recommendations and distribution for cultural activities. 连接艺术与观众的智能票务平台，为文化活动提供智能推荐与分发。
BetaZero-to-one toolkits. Quick-start guides and templates for beginners in any field. 零起步工具集。为新手提供快速入门指南和模板。
LiveInfinite creation space. Multi-dimensional knowledge management and decision support. 无限创造空间。多维度知识管理和决策支持系统。
Coming SoonTime-value marketplace. Knowledge and service exchange based on time economics. 时间商城。基于时间价值的知识和服务交易平台。
Coming SoonAgent-ready web infrastructure. Build and deploy static pages with no code. 开放页面云。快速构建和部署静态页面的无代码平台。
LiveA research-driven AI company with offices in the UK and China. 研究驱动的 AI 公司，在英国和中国设有办公室。
Soodoo Ltd
London, UK
Global Operations 全球业务
时代数渡
Shanghai, China
Asia-Pacific Operations 亚太业务