Category
Technology
RAG-as-a-Service and the Modern State: Context Bundling, Derived Pages, and Grounded Truth
How Wikantik evolved from a human wiki to an agentic RAG-as-a-Service platform featuring context bundling, Apache Tika ingestion, and zero answer synthesis.
First-Class AI Citizens: Designing Dual Model Context Protocol Architectures
How Wikantik partitioned its agent interface into dual MCP servers: /knowledge-mcp for lightweight RAG retrieval and /wikantik-admin-mcp for verified writes.
The Two Graphs: Disentangling Page Graphs from Semantic Knowledge Graphs
How Wikantik separated document hyperlink topology (the Page Graph) from semantic entity ontologies (the Knowledge Graph) to eliminate architectural confusion.
Evaluating GraphRAG: Benchmarks and the Removal of Knowledge Graph Reranking
Engineering postmortem on GraphRAG: why page-level entity co-mention graphs produced zero net retrieval lift across 40 evaluation queries and why we removed 2,464 lines of code.
Building the Hybrid Retrieval Engine: BM25 Meets pgvector and Reciprocal Rank Fusion
How Wikantik unified Lucene BM25 keyword search with PostgreSQL pgvector dense embeddings using heading-aware chunking and Reciprocal Rank Fusion.
Decoupling the Legacy Web Tier: The React SPA Migration and Database-Backed RBAC
How Wikantik replaced server-side JSP tag libraries with a Vite + React SPA, stateless REST endpoints, PostgreSQL RBAC, and Pac4j SSO.
From JSPWiki to Wikantik: The Jakarta EE 10 Migration
How Wikantik migrated 261 files from legacy javax.* to Jakarta EE 10, replaced Stripes test mocks with Mockito, and updated to Tomcat 10/11.