Qdrant
A vector retrieval database supporting dense and sparse vectors, payload filtering, and hybrid search. It can serve as a RAG retrieval layer in local containers, self-hosted clusters, and edge deployments.
SEARCH OPNFTR
Search published projects, source resources and knowledge articles.
26 results for “RAG”Clear search
A vector retrieval database supporting dense and sparse vectors, payload filtering, and hybrid search. It can serve as a RAG retrieval layer in local containers, self-hosted clusters, and edge deployments.
An application framework for connecting private data with large language models, covering data loading, indexing, retrieval, RAG flows, and Agent or Workflow components. It is suited to enterprise knowledge retrieval and data-intensive LLM applications.
An open-source library for sentence and document embeddings, semantic similarity, and reranking. It is suitable for semantic search, clustering, deduplication, and RAG retrieval evaluation workflows.
An open-source framework for search, RAG, agents, and document processing pipelines. It provides component-based connectors, retrieval, generation, and evaluation for enterprise AI applications managed as explicit pipelines.
A safe feedback loop that turns reviewed user signals into regression tests, targeted fixes, and reversible releases.
A layered evaluation and tracing approach that turns AI quality, latency, and cost into actionable release signals.
A traceable multimodal pipeline with input normalization, asynchronous work, quality sampling, and privacy-aware deletion.
A practical, repeatable approach to chunking, hybrid retrieval, evaluation, and authorization-aware vector search.
A review-first AI coding workflow that keeps generated changes small, testable, traceable, and safe.
A practical method for versioned prompts with explicit contracts, evaluation samples, and recoverable failure paths.
A practical architecture for chunking, hybrid retrieval, citations, access control, evaluation, and cost management in enterprise RAG.
Codex 与 LLM Wiki 案例,说明如何在 Obsidian 中搭建 AI 知识库、整理目录、引用和更新流程。
Codex Chrome 浏览器插件案例,说明如何让 AI 控制浏览器页面、执行网页任务并保持安全边界。
Codex 连接第三方 API 的入门说明,比较手动配置、Codex++、CCX 与 CC Switch 三种接入方式,并提示账号、密钥、账单和隐私风险。
从可追溯文档管线开始,定义内容、版本、权限和删除语义。
选择 embedding、索引距离和批处理策略,建立可重建的向量索引。
组合关键词、向量和过滤条件,把候选相关性与权限边界分开处理。
用可追溯问题集检查召回覆盖、引用准确和无依据回答。
把租户权限、文档状态和更新时间作为检索的一等字段。
设计可观测的索引发布流程,避免半成品索引和缓存污染答案。
用受限查询改写和证据阈值处理口语问题、无结果和不确定回答。
用 SQLite FTS 和确定性上下文拼装一个无需外部服务的可复现 RAG。
构建新索引、执行抽样检查,再用版本指针完成无损切换。
用来源 id、权限标签和期望证据回放检索变化,区分召回回退与生成问题。
把检索结果变成可追溯上下文,避免把无权限文档送给模型。
用 SQLite FTS 验证角色过滤、来源引用、无结果和删除语义。