Fidelity Chauke

@fidschau.bsky.social

41 Followers 150 Following

Android engineer. Playing with Koog. Building ntivo

Picture of myself In front of a screen with an ai generated cloud
A vertical flow diagram titled "Ntivo architecture" showing six layers connected by arrows from top to bottom. The first layer, Sources, contains five components in teal: GitHub (code repos), Jira (tickets and epics), Confluence (specs and ADRs), Figma (screens and flows), and Datadog (alerts and traces). The second layer, Ingest, contains three purple components: Tree-sitter (AST to chunks), Ingestion engine (extract, fingerprint, route), and Gemini embedder (gemini-embedding-001, BYOK). The third layer, Store, contains two amber components side by side: Neo4j (knowledge graph, explicit relationships) and Qdrant (vector store, semantic similarity). The fourth layer, Agent, is a single wide coral component spanning the full width: the Koog AI agent, which accepts a bring-your-own-key LLM from Gemini, OpenAI, Anthropic, and others, and exposes three tools — graph_traverse, vector_search, and code_grep — before synthesising an answer with source attribution. The fifth layer is the Ktor API server in gray, exposing REST, MCP, and A2A protocols. The sixth and final layer, Consume, contains three blue components: Compose for Web (developer console), MCP clients (Claude Code and Junie), and A2A agents (planning and triage bots).
Ntivo architecture diagram: six layers flowing top to bottom. Sources (GitHub, Jira, Confluence, Figma, Datadog) feed into an ingestion layer (Tree-sitter, Gemini embedder), stored in Neo4j and Qdrant, queried by a Koog AI agent, served via Ktor API, consumed by web UI, MCP clients, and A2A agents.