it-ops/llm-automation-docs-and-remediation-engine

Fork 0

Files

Daniele Viti 8d11321835

Build / Code Quality Checks (push) Successful in 17m40s

Details

Build / Build & Push Docker Images (frontend) (push) Failing after 2m22s

Details

Build / Build & Push Docker Images (worker) (push) Successful in 20m15s

Details

Build / Build & Push Docker Images (chat) (push) Failing after 45m14s

Details

Build / Build & Push Docker Images (api) (push) Failing after 45m14s

Details

Update SCHEME.md

2025-10-28 11:42:17 +00:00

32 KiB

Raw Blame History

📚 Automated Infrastructure Documentation System

Sistema automatizzato per la generazione e mantenimento della documentazione tecnica dell'infrastruttura aziendale tramite LLM locale con validazione umana e pubblicazione GitOps.

🎯 Overview

Sistema progettato per automatizzare la creazione e l'aggiornamento della documentazione tecnica di sistemi infrastrutturali complessi (VMware, Kubernetes, Linux, Cisco, ecc.) utilizzando un Large Language Model locale (Qwen).

Caratteristiche Principali

✅ Raccolta dati asincrona da molteplici sistemi infrastrutturali
✅ Isolamento di sicurezza: LLM non accede mai ai sistemi live
✅ Change Detection: Documentazione generata solo su modifiche rilevate
✅ Redis Cache per storage dati e performance
✅ LLM locale on-premise (Qwen) tramite MCP Server
✅ Human-in-the-loop validation con workflow GitOps
✅ CI/CD automatizzato per pubblicazione

🏗️ Architettura

Il sistema è suddiviso in 3 flussi principali:

Raccolta Dati (Background): Connettori interrogano periodicamente i sistemi infrastrutturali tramite API e aggiornano Redis
Change Detection: Sistema di rilevamento modifiche che attiva la generazione documentazione solo quando necessario
Generazione e Pubblicazione (Triggered): LLM locale (Qwen) genera markdown leggendo da Redis, seguito da review umana e deploy automatico

Principio di Sicurezza: L'LLM non ha mai accesso diretto ai sistemi infrastrutturali. Tutti i dati sono letti da Redis.

Principio di Efficienza: La documentazione viene generata solo quando il sistema rileva modifiche nella configurazione infrastrutturale.

📊 Schema Architetturale

Management View

Schema semplificato per presentazioni executive e management.

graph TB
    %% Styling
    classDef infrastructure fill:#e1f5ff,stroke:#01579b,stroke-width:3px,color:#333
    classDef cache fill:#f3e5f5,stroke:#4a148c,stroke-width:3px,color:#333
    classDef change fill:#fff3e0,stroke:#e65100,stroke-width:3px,color:#333
    classDef llm fill:#e8f5e9,stroke:#1b5e20,stroke-width:3px,color:#333
    classDef git fill:#fce4ec,stroke:#880e4f,stroke-width:3px,color:#333
    classDef human fill:#fff9c4,stroke:#f57f17,stroke-width:3px,color:#333
    
    %% ========================================
    %% FLUSSO 1: RACCOLTA DATI (Background)
    %% ========================================
    
    INFRA[("🏢 SISTEMI<br/>INFRASTRUTTURALI<br/><br/>VMware | K8s | Linux | Cisco")]:::infrastructure
    
    CONN["🔌 CONNETTORI<br/>Polling Automatico"]:::infrastructure
    
    REDIS[("💾 REDIS CACHE<br/>Configurazione<br/>Infrastruttura")]:::cache
    
    INFRA -->|"API Polling<br/>Continuo"| CONN
    CONN -->|"Update<br/>Configurazione"| REDIS
    
    %% ========================================
    %% CHANGE DETECTION
    %% ========================================
    
    CHANGE["🔍 CHANGE DETECTOR<br/>Rileva Modifiche<br/>Configurazione"]:::change
    
    REDIS -->|"Monitor<br/>Changes"| CHANGE
    
    %% ========================================
    %% FLUSSO 2: GENERAZIONE DOCUMENTAZIONE (Triggered)
    %% ========================================
    
    TRIGGER["⚡ TRIGGER<br/>Solo se modifiche"]:::change
    
    USER["👤 UTENTE<br/>Richiesta Manuale"]:::human
    
    LLM["🤖 LLM ENGINE<br/>Qwen (Locale)"]:::llm
    
    MCP["🔧 MCP SERVER<br/>API Control Platform"]:::llm
    
    DOC["📄 DOCUMENTO<br/>Markdown Generato"]:::llm
    
    CHANGE -->|"Modifiche<br/>Rilevate"| TRIGGER
    USER -.->|"Opzionale"| TRIGGER
    
    TRIGGER -->|"Avvia<br/>Generazione"| LLM
    LLM -->|"Tool Call"| MCP
    MCP -->|"Query"| REDIS
    REDIS -->|"Dati Config"| MCP
    MCP -->|"Context"| LLM
    LLM -->|"Genera"| DOC
    
    %% ========================================
    %% FLUSSO 3: VALIDAZIONE E PUBBLICAZIONE
    %% ========================================
    
    GIT["📦 GITLAB<br/>Repository"]:::git
    
    PR["🔀 PULL REQUEST<br/>Review Automatica"]:::git
    
    TECH["👨‍💼 TEAM TECNICO<br/>Validazione Umana"]:::human
    
    PIPELINE["⚡ CI/CD PIPELINE<br/>GitLab Runner"]:::git
    
    MKDOCS["📚 MKDOCS<br/>Static Site Generator"]:::git
    
    WEB["🌐 DOCUMENTAZIONE<br/>GitLab Pages<br/>(Pubblicata)"]:::git
    
    DOC -->|"Push +<br/>Branch"| GIT
    GIT -->|"Crea"| PR
    PR -->|"Notifica"| TECH
    TECH -->|"Approva +<br/>Merge"| GIT
    GIT -->|"Trigger"| PIPELINE
    PIPELINE -->|"Build"| MKDOCS
    MKDOCS -->|"Deploy"| WEB
    
    %% ========================================
    %% ANNOTAZIONI
    %% ========================================
    
    SECURITY["🔒 SICUREZZA<br/>LLM isolato dai sistemi live"]:::human
    EFFICIENCY["⚡ EFFICIENZA<br/>Doc generata solo<br/>su modifiche"]:::change
    
    LLM -.->|"NESSUN<br/>ACCESSO"| INFRA
    
    SECURITY -.-> LLM
    EFFICIENCY -.-> CHANGE

🔧 Schema Tecnico

Implementation View

Schema dettagliato per il team tecnico con specifiche implementative.

graph TB
    %% Styling tecnico
    classDef infra fill:#e1f5ff,stroke:#01579b,stroke-width:2px,color:#333,font-size:11px
    classDef connector fill:#e3f2fd,stroke:#1565c0,stroke-width:2px,color:#333,font-size:11px
    classDef cache fill:#f3e5f5,stroke:#4a148c,stroke-width:2px,color:#333,font-size:11px
    classDef change fill:#fff3e0,stroke:#e65100,stroke-width:2px,color:#333,font-size:11px
    classDef llm fill:#e8f5e9,stroke:#1b5e20,stroke-width:2px,color:#333,font-size:11px
    classDef git fill:#fce4ec,stroke:#880e4f,stroke-width:2px,color:#333,font-size:11px
    classDef monitor fill:#fff8e1,stroke:#f57f17,stroke-width:2px,color:#333,font-size:11px
    
    %% =====================================
    %% LAYER 1: SISTEMI SORGENTE
    %% =====================================
    
    subgraph SOURCES["🏢 INFRASTRUCTURE SOURCES"]
        VCENTER["VMware vCenter<br/>API: vSphere REST 7.0+<br/>Port: 443/HTTPS<br/>Auth: API Token"]:::infra
        K8S_API["Kubernetes API<br/>API: v1.28+<br/>Port: 6443/HTTPS<br/>Auth: ServiceAccount + RBAC"]:::infra
        LINUX["Linux Servers<br/>Protocol: SSH/Ansible<br/>Port: 22<br/>Auth: SSH Keys"]:::infra
        CISCO["Cisco Devices<br/>Protocol: NETCONF/RESTCONF<br/>Port: 830/443<br/>Auth: AAA"]:::infra
    end
    
    %% =====================================
    %% LAYER 2: CONNETTORI
    %% =====================================
    
    subgraph CONNECTORS["🔌 DATA COLLECTORS (Python/Go)"]
        CONN_VM["VMware Collector<br/>Lang: Python 3.11<br/>Lib: pyvmomi<br/>Schedule: */15 * * * *<br/>Output: JSON → Redis"]:::connector
        
        CONN_K8S["K8s Collector<br/>Lang: Python 3.11<br/>Lib: kubernetes-client<br/>Schedule: */5 * * * *<br/>Resources: pods,svc,ing,deploy"]:::connector
        
        CONN_LNX["Linux Collector<br/>Lang: Python 3.11<br/>Lib: paramiko/ansible<br/>Schedule: */30 * * * *<br/>Data: sysinfo,packages,services"]:::connector
        
        CONN_CSC["Cisco Collector<br/>Lang: Python 3.11<br/>Lib: ncclient<br/>Schedule: */30 * * * *<br/>Data: interfaces,routing,vlans"]:::connector
    end
    
    VCENTER -->|"GET /api/vcenter/vm"| CONN_VM
    K8S_API -->|"kubectl proxy<br/>API calls"| CONN_K8S
    LINUX -->|"SSH batch<br/>commands"| CONN_LNX
    CISCO -->|"NETCONF<br/>get-config"| CONN_CSC
    
    %% =====================================
    %% LAYER 3: REDIS STORAGE
    %% =====================================
    
    subgraph STORAGE["💾 REDIS CLUSTER"]
        REDIS_CLUSTER["Redis Cluster<br/>Mode: Cluster (6 nodes)<br/>Port: 6379<br/>Persistence: RDB + AOF<br/>Memory: 64GB<br/>Eviction: allkeys-lru"]:::cache
        
        REDIS_KEYS["Key Structure:<br/>• vmware:vcenter-id:vms:hash<br/>• k8s:cluster:namespace:resource:hash<br/>• linux:hostname:info:hash<br/>• cisco:device-id:config:hash<br/>• changelog:timestamp:diff<br/>TTL: 30d for data, 90d for changelog"]:::cache
    end
    
    CONN_VM -->|"HSET/HMSET<br/>+ Hash Storage"| REDIS_CLUSTER
    CONN_K8S -->|"HSET/HMSET<br/>+ Hash Storage"| REDIS_CLUSTER
    CONN_LNX -->|"HSET/HMSET<br/>+ Hash Storage"| REDIS_CLUSTER
    CONN_CSC -->|"HSET/HMSET<br/>+ Hash Storage"| REDIS_CLUSTER
    
    REDIS_CLUSTER --> REDIS_KEYS
    
    %% =====================================
    %% LAYER 4: CHANGE DETECTION
    %% =====================================
    
    subgraph CHANGE_DETECTION["🔍 CHANGE DETECTION SYSTEM"]
        DETECTOR["Change Detector Service<br/>Lang: Python 3.11<br/>Lib: redis-py<br/>Algorithm: Hash comparison<br/>Check interval: */5 * * * *"]:::change
        
        DIFF_ENGINE["Diff Engine<br/>• Deep object comparison<br/>• JSON diff generation<br/>• Change classification<br/>• Severity assessment"]:::change
        
        CHANGE_LOG["Change Log Store<br/>Key: changelog:*<br/>Data: diff JSON + metadata<br/>Indexed by: timestamp, resource"]:::change
        
        NOTIFIER["Change Notifier<br/>• Webhook triggers<br/>• Slack notifications<br/>• Event emission<br/>Target: LLM trigger"]:::change
    end
    
    REDIS_CLUSTER -->|"Monitor<br/>key changes"| DETECTOR
    DETECTOR --> DIFF_ENGINE
    DIFF_ENGINE -->|"Store diff"| CHANGE_LOG
    CHANGE_LOG --> REDIS_CLUSTER
    DIFF_ENGINE -->|"Notify if<br/>significant"| NOTIFIER
    
    %% =====================================
    %% LAYER 5: LLM TRIGGER & GENERATION
    %% =====================================
    
    subgraph TRIGGER_SYSTEM["⚡ TRIGGER SYSTEM"]
        TRIGGER_SVC["Trigger Service<br/>Lang: Python 3.11<br/>Listen: Webhook + Redis Pub/Sub<br/>Debounce: 5 min<br/>Batch: multiple changes"]:::change
        
        QUEUE["Generation Queue<br/>Type: Redis List<br/>Priority: High/Medium/Low<br/>Processing: FIFO"]:::change
    end
    
    NOTIFIER -->|"Trigger event"| TRIGGER_SVC
    TRIGGER_SVC -->|"Enqueue<br/>generation task"| QUEUE
    
    subgraph LLM_LAYER["🤖 AI GENERATION LAYER"]
        LLM_ENGINE["LLM Engine<br/>Model: Qwen (Locale)<br/>API: Ollama/vLLM/LM Studio<br/>Port: 11434<br/>Temp: 0.3<br/>Max Tokens: 4096<br/>Timeout: 120s"]:::llm
        
        MCP_SERVER["MCP Server<br/>Lang: TypeScript/Node.js<br/>Port: 3000<br/>Protocol: JSON-RPC 2.0<br/>Auth: JWT tokens"]:::llm
        
        MCP_TOOLS["MCP Tools:<br/>• getVMwareInventory(vcenter)<br/>• getK8sResources(cluster,ns,type)<br/>• getLinuxSystemInfo(hostname)<br/>• getCiscoConfig(device,section)<br/>• getChangelog(start,end,resource)<br/>Return: JSON + Metadata"]:::llm
    end
    
    QUEUE -->|"Dequeue<br/>task"| LLM_ENGINE
    
    LLM_ENGINE <-->|"Tool calls<br/>JSON-RPC"| MCP_SERVER
    MCP_SERVER --> MCP_TOOLS
    
    MCP_TOOLS -->|"HGETALL/MGET<br/>Read data"| REDIS_CLUSTER
    REDIS_CLUSTER -->|"Config data<br/>+ Changelog"| MCP_TOOLS
    MCP_TOOLS -->|"Structured Data<br/>+ Context"| LLM_ENGINE
    
    subgraph OUTPUT["📝 DOCUMENT GENERATION"]
        TEMPLATE["Template Engine<br/>Format: Jinja2<br/>Templates: markdown/*.j2<br/>Variables: from LLM"]:::llm
        
        MARKDOWN["Markdown Output<br/>Format: CommonMark<br/>Metadata: YAML frontmatter<br/>Change summary included<br/>Assets: diagrams in mermaid"]:::llm
        
        VALIDATOR["Doc Validator<br/>• Markdown linting<br/>• Link checking<br/>• Schema validation<br/>• Change verification"]:::llm
    end
    
    LLM_ENGINE --> TEMPLATE
    TEMPLATE --> MARKDOWN
    MARKDOWN --> VALIDATOR
    
    %% =====================================
    %% LAYER 6: GITOPS
    %% =====================================
    
    subgraph GITOPS["🔄 GITOPS WORKFLOW"]
        GIT_REPO["GitLab Repository<br/>URL: gitlab.com/docs/infra<br/>Branch strategy: main + feature/*<br/>Protected: main (require approval)"]:::git
        
        GIT_API["GitLab API<br/>API: v4<br/>Auth: Project Access Token<br/>Permissions: api, write_repo"]:::git
        
        PR_AUTO["Automated PR Creator<br/>Lang: Python 3.11<br/>Lib: python-gitlab<br/>Template: .gitlab/merge_request.md<br/>Include: change summary"]:::git
    end
    
    VALIDATOR -->|"git add/commit/push"| GIT_REPO
    GIT_REPO <--> GIT_API
    GIT_API --> PR_AUTO
    
    REVIEWER["👨‍💼 Technical Reviewer<br/>Role: Maintainer/Owner<br/>Review: diff + validation<br/>Check: change correlation<br/>Approve: required (min 1)"]:::monitor
    
    PR_AUTO -->|"Notification<br/>Email + Slack"| REVIEWER
    REVIEWER -->|"Merge to main"| GIT_REPO
    
    %% =====================================
    %% LAYER 7: CI/CD & PUBLISH
    %% =====================================
    
    subgraph CICD["⚡ CI/CD PIPELINE"]
        GITLAB_CI["GitLab CI/CD<br/>Runner: docker<br/>Image: python:3.11-alpine<br/>Stages: build, test, deploy"]:::git
        
        PIPELINE_JOBS["Pipeline Jobs:<br/>1. lint (markdownlint-cli)<br/>2. build (mkdocs build)<br/>3. test (link-checker)<br/>4. deploy (rsync/s3)"]:::git
        
        MKDOCS_CFG["MkDocs Config<br/>Theme: material<br/>Plugins: search, tags, mermaid<br/>Extensions: admonition, codehilite"]:::git
    end
    
    GIT_REPO -->|"on: push to main<br/>Webhook trigger"| GITLAB_CI
    GITLAB_CI --> PIPELINE_JOBS
    PIPELINE_JOBS --> MKDOCS_CFG
    
    subgraph PUBLISH["🌐 PUBLICATION"]
        STATIC_SITE["Static Site<br/>Generator: MkDocs<br/>Output: HTML/CSS/JS<br/>Assets: optimized images"]:::git
        
        CDN["GitLab Pages / S3 + CloudFront<br/>URL: docs.company.com<br/>SSL: Let's Encrypt<br/>Cache: 1h"]:::git
        
        SEARCH["Search Index<br/>Engine: Algolia/Meilisearch<br/>Update: on publish<br/>API: REST"]:::git
    end
    
    MKDOCS_CFG -->|"mkdocs build<br/>--strict"| STATIC_SITE
    STATIC_SITE --> CDN
    STATIC_SITE --> SEARCH
    
    %% =====================================
    %% LAYER 8: MONITORING & OBSERVABILITY
    %% =====================================
    
    subgraph OBSERVABILITY["📊 MONITORING & LOGGING"]
        PROMETHEUS["Prometheus<br/>Metrics: collector updates, changes detected<br/>Scrape: 30s<br/>Retention: 15d"]:::monitor
        
        GRAFANA["Grafana Dashboards<br/>• Collector status<br/>• Redis performance<br/>• Change detection rate<br/>• LLM response times<br/>• Pipeline success rate"]:::monitor
        
        ELK["ELK Stack<br/>Logs: all components<br/>Index: daily rotation<br/>Retention: 30d"]:::monitor
        
        ALERTS["Alerting<br/>• Collector failures<br/>• Redis issues<br/>• Change detection errors<br/>• Pipeline failures<br/>Channel: Slack + PagerDuty"]:::monitor
    end
    
    CONN_VM -.->|"metrics"| PROMETHEUS
    CONN_K8S -.->|"metrics"| PROMETHEUS
    REDIS_CLUSTER -.->|"metrics"| PROMETHEUS
    DETECTOR -.->|"metrics"| PROMETHEUS
    MCP_SERVER -.->|"metrics"| PROMETHEUS
    GITLAB_CI -.->|"metrics"| PROMETHEUS
    
    PROMETHEUS --> GRAFANA
    
    CONN_VM -.->|"logs"| ELK
    DETECTOR -.->|"logs"| ELK
    MCP_SERVER -.->|"logs"| ELK
    GITLAB_CI -.->|"logs"| ELK
    
    GRAFANA --> ALERTS
    
    %% =====================================
    %% SECURITY & EFFICIENCY ANNOTATIONS
    %% =====================================
    
    SEC1["🔒 SECURITY:<br/>• All APIs use TLS 1.3<br/>• Secrets in Vault/K8s Secrets<br/>• Network: private VPC<br/>• LLM has NO direct access"]:::monitor
    
    SEC2["🔐 AUTHENTICATION:<br/>• API Tokens rotated 90d<br/>• RBAC enforced<br/>• Audit logs enabled<br/>• MFA required for Git"]:::monitor
    
    EFF1["⚡ EFFICIENCY:<br/>• Doc generation only on changes<br/>• Debounce prevents spam<br/>• Hash-based change detection<br/>• Batch processing"]:::change
    
    SEC1 -.-> MCP_SERVER
    SEC2 -.-> GIT_REPO
    EFF1 -.-> DETECTOR

💬 Sistema RAG Conversazionale

Interrogazione Documentazione con AI

Sistema per "parlare" con la documentazione utilizzando Retrieval Augmented Generation (RAG). Permette agli utenti di porre domande in linguaggio naturale e ricevere risposte accurate basate sulla documentazione, con citazioni delle fonti.

Caratteristiche Principali

✅ Semantic Search: Ricerca vettoriale per comprendere l'intento della query
✅ Scalabilità: Gestione di grandi volumi di documentazione (100k+ documenti)
✅ Performance: Risposte in <3 secondi con caching intelligente
✅ Accuratezza: Re-ranking e source attribution per risposte precise
✅ LLM Locale: Qwen on-premise per privacy e controllo

Schema RAG - Management View

graph TB
    %% Styling
    classDef docs fill:#e3f2fd,stroke:#1565c0,stroke-width:3px,color:#333
    classDef process fill:#f3e5f5,stroke:#4a148c,stroke-width:3px,color:#333
    classDef vector fill:#fff3e0,stroke:#e65100,stroke-width:3px,color:#333
    classDef llm fill:#e8f5e9,stroke:#1b5e20,stroke-width:3px,color:#333
    classDef user fill:#fff9c4,stroke:#f57f17,stroke-width:3px,color:#333
    classDef cache fill:#fce4ec,stroke:#880e4f,stroke-width:3px,color:#333

    %% ========================================
    %% INGESTION PIPELINE (Offline)
    %% ========================================

    subgraph INGESTION["📚 INGESTION PIPELINE (Offline Process)"]
        DOCS["📄 DOCUMENTAZIONE<br/>MkDocs Output<br/>Markdown Files"]:::docs

        CHUNKER["✂️ DOCUMENT CHUNKER<br/>Split & Overlap<br/>Metadata Extraction"]:::process

        EMBEDDER["🧠 EMBEDDING MODEL<br/>Text → Vectors<br/>Dimensione: 768/1024"]:::process

        VECTORDB[("🗄️ VECTOR DATABASE<br/>Qdrant/Milvus<br/>Sharded & Replicated")]:::vector
    end

    DOCS -->|"Parse<br/>Markdown"| CHUNKER
    CHUNKER -->|"Text Chunks<br/>+ Metadata"| EMBEDDER
    EMBEDDER -->|"Store<br/>Embeddings"| VECTORDB

    %% ========================================
    %% QUERY PIPELINE (Real-time)
    %% ========================================

    subgraph QUERY["💬 QUERY PIPELINE (Real-time)"]
        USER["👤 UTENTE<br/>Domanda/Query"]:::user

        QUERY_EMBED["🧠 QUERY EMBEDDING<br/>Query → Vector"]:::process

        SEARCH["🔍 SEMANTIC SEARCH<br/>Vector Similarity<br/>Top-K Results"]:::vector

        RERANK["📊 RE-RANKING<br/>Context Scoring<br/>Relevance Filter"]:::process

        CONTEXT["📋 CONTEXT BUILDER<br/>Assemble Chunks<br/>Add Metadata"]:::process
    end

    USER -->|"Natural Language<br/>Question"| QUERY_EMBED
    QUERY_EMBED -->|"Query Vector"| SEARCH
    SEARCH -->|"Search"| VECTORDB
    VECTORDB -->|"Top-K Chunks<br/>+ Scores"| SEARCH
    SEARCH -->|"Initial Results"| RERANK
    RERANK -->|"Filtered<br/>Chunks"| CONTEXT

    %% ========================================
    %% GENERATION (LLM)
    %% ========================================

    subgraph GENERATION["🤖 ANSWER GENERATION"]
        LLM_RAG["🤖 LLM ENGINE<br/>Qwen (Locale)<br/>+ RAG Context"]:::llm

        ANSWER["💡 RISPOSTA<br/>Generated Answer<br/>+ Source Citations"]:::llm
    end

    CONTEXT -->|"Context<br/>+ Sources"| LLM_RAG
    LLM_RAG -->|"Generate"| ANSWER
    ANSWER -->|"Display"| USER

    %% ========================================
    %% CACHING & OPTIMIZATION
    %% ========================================

    CACHE[("💾 REDIS CACHE<br/>Query Cache<br/>Embedding Cache")]:::cache

    QUERY_EMBED -.->|"Check Cache"| CACHE
    CACHE -.->|"Cached<br/>Embedding"| SEARCH

    SEARCH -.->|"Cache<br/>Results"| CACHE

    %% ========================================
    %% SCALING & UPDATE
    %% ========================================

    UPDATE["🔄 INCREMENTAL UPDATE<br/>On Doc Changes<br/>Auto Re-index"]:::docs

    DOCS -.->|"Doc Updated"| UPDATE
    UPDATE -.->|"Re-process<br/>Changed Docs"| CHUNKER

    %% ========================================
    %% ANNOTATIONS
    %% ========================================

    SCALE["📈 SCALABILITÀ<br/>• Vector DB sharding<br/>• Horizontal scaling<br/>• Load balancing"]:::vector

    PERF["⚡ PERFORMANCE<br/>• Query cache<br/>• Embedding cache<br/>• Async processing"]:::cache

    QUALITY["✅ QUALITY<br/>• Re-ranking<br/>• Relevance scoring<br/>• Source citations"]:::process

    SCALE -.-> VECTORDB
    PERF -.-> CACHE
    QUALITY -.-> RERANK

Schema RAG - Technical View

graph TB
    %% Styling
    classDef docs fill:#e3f2fd,stroke:#1565c0,stroke-width:2px,color:#333,font-size:11px
    classDef process fill:#f3e5f5,stroke:#4a148c,stroke-width:2px,color:#333,font-size:11px
    classDef vector fill:#fff3e0,stroke:#e65100,stroke-width:2px,color:#333,font-size:11px
    classDef llm fill:#e8f5e9,stroke:#1b5e20,stroke-width:2px,color:#333,font-size:11px
    classDef user fill:#fff9c4,stroke:#f57f17,stroke-width:2px,color:#333,font-size:11px
    classDef cache fill:#fce4ec,stroke:#880e4f,stroke-width:2px,color:#333,font-size:11px
    classDef monitor fill:#fff8e1,stroke:#f57f17,stroke-width:2px,color:#333,font-size:11px

    %% =====================================
    %% LAYER 1: DOCUMENTATION SOURCE
    %% =====================================

    subgraph DOCSOURCE["📚 DOCUMENTATION SOURCE"]
        MKDOCS_OUT["MkDocs Static Site<br/>Path: /site/<br/>Format: HTML + Markdown<br/>Assets: images, diagrams<br/>Update: on Git merge"]:::docs

        DOC_WATCHER["Document Watcher<br/>Lang: Python 3.11<br/>Lib: watchdog<br/>Trigger: file system events<br/>Debounce: 30s"]:::docs

        DOC_PARSER["Document Parser<br/>HTML → Plain Text<br/>Preserve structure<br/>Extract metadata<br/>Clean formatting"]:::docs
    end

    MKDOCS_OUT --> DOC_WATCHER
    DOC_WATCHER -->|"New/Modified<br/>Docs"| DOC_PARSER

    %% =====================================
    %% LAYER 2: CHUNKING STRATEGY
    %% =====================================

    subgraph CHUNKING["✂️ INTELLIGENT CHUNKING"]
        CHUNK_ENGINE["Chunking Engine<br/>Lang: Python 3.11<br/>Lib: langchain/llama-index<br/>Strategy: Recursive Character"]:::process

        CHUNK_CONFIG["Chunking Config:<br/>• Chunk Size: 512 tokens<br/>• Overlap: 128 tokens<br/>• Separators: \\n\\n, \\n, . , ' '<br/>• Min chunk: 100 tokens<br/>• Max chunk: 1024 tokens"]:::process

        METADATA_EXTRACTOR["Metadata Extractor<br/>Extract:<br/>• Document title<br/>• Section headers<br/>• Tags/keywords<br/>• Creation date<br/>• File path<br/>• Doc type"]:::process
    end

    DOC_PARSER -->|"Parsed Text"| CHUNK_ENGINE
    CHUNK_ENGINE --> CHUNK_CONFIG
    CHUNK_ENGINE --> METADATA_EXTRACTOR

    %% =====================================
    %% LAYER 3: EMBEDDING GENERATION
    %% =====================================

    subgraph EMBEDDING["🧠 EMBEDDING GENERATION"]
        EMBED_MODEL["Embedding Model<br/>Model: all-MiniLM-L6-v2 / BGE-M3<br/>Dim: 384/768/1024<br/>API: sentence-transformers<br/>Batch size: 32<br/>GPU: CUDA acceleration"]:::process

        EMBED_CACHE["Embedding Cache<br/>Type: Redis Hash<br/>Key: hash(text)<br/>TTL: 30d<br/>Hit rate target: >80%"]:::cache

        EMBED_QUEUE["Processing Queue<br/>Type: Redis List<br/>Workers: 4-8<br/>Rate: 100 chunks/s<br/>Retry: 3 attempts"]:::process
    end

    METADATA_EXTRACTOR -->|"Chunks<br/>+ Metadata"| EMBED_QUEUE
    EMBED_QUEUE --> EMBED_MODEL
    EMBED_MODEL <-.->|"Cache<br/>Check/Store"| EMBED_CACHE

    %% =====================================
    %% LAYER 4: VECTOR DATABASE
    %% =====================================

    subgraph VECTORDB["🗄️ VECTOR DATABASE CLUSTER"]
        QDRANT["Qdrant Cluster<br/>Version: 1.7+<br/>Nodes: 3-6 (replicated)<br/>Shards: auto per collection<br/>Port: 6333/6334"]:::vector

        COLLECTIONS["Collections:<br/>• docs_main (dim: 768)<br/>• docs_code (dim: 768)<br/>• docs_api (dim: 768)<br/>Distance: Cosine<br/>Index: HNSW (M=16, ef=100)"]:::vector

        SHARD_STRATEGY["Sharding Strategy:<br/>• Auto-sharding enabled<br/>• Shard size: 100k vectors<br/>• Replication factor: 2<br/>• Load balancing: Round-robin"]:::vector
    end

    EMBED_MODEL -->|"Store<br/>Vectors"| QDRANT
    QDRANT --> COLLECTIONS
    QDRANT --> SHARD_STRATEGY

    %% =====================================
    %% LAYER 5: QUERY PROCESSING
    %% =====================================

    subgraph QUERYPROC["💬 QUERY PROCESSING PIPELINE"]
        USER_INPUT["User Input<br/>Interface: Web UI / API<br/>Auth: JWT tokens<br/>Rate limit: 20 req/min<br/>Timeout: 30s"]:::user

        QUERY_PREPROCESS["Query Preprocessor<br/>• Spelling correction<br/>• Intent detection<br/>• Query expansion<br/>• Language detection"]:::process

        QUERY_EMBEDDER["Query Embedder<br/>Same model as docs<br/>Cache: Redis<br/>Latency: <50ms"]:::process

        HYBRID_SEARCH["Hybrid Search<br/>1. Vector search (semantic)<br/>2. Keyword search (BM25)<br/>3. Fusion: RRF algorithm<br/>Top-K: 20 initial results"]:::vector
    end

    USER_INPUT -->|"Natural<br/>Language"| QUERY_PREPROCESS
    QUERY_PREPROCESS --> QUERY_EMBEDDER
    QUERY_EMBEDDER <-.->|"Cache"| EMBED_CACHE
    QUERY_EMBEDDER -->|"Query<br/>Vector"| HYBRID_SEARCH
    HYBRID_SEARCH -->|"Search"| QDRANT

    %% =====================================
    %% LAYER 6: RE-RANKING & FILTERING
    %% =====================================

    subgraph RERANK["📊 RE-RANKING & FILTERING"]
        RERANKER["Cross-Encoder Re-ranker<br/>Model: ms-marco-MiniLM<br/>Purpose: Fine-grained relevance<br/>Process: Top-20 → Top-5<br/>Latency: 100-200ms"]:::process

        FILTER_ENGINE["Filter Engine<br/>• Relevance threshold: >0.7<br/>• Deduplication<br/>• Diversity scoring<br/>• Metadata filtering"]:::process

        CONTEXT_BUILDER["Context Builder<br/>• Assemble top chunks<br/>• Add source citations<br/>• Format for LLM<br/>• Max context: 4k tokens"]:::process
    end

    QDRANT -->|"Top-K<br/>Results"| RERANKER
    RERANKER --> FILTER_ENGINE
    FILTER_ENGINE --> CONTEXT_BUILDER

    %% =====================================
    %% LAYER 7: LLM GENERATION
    %% =====================================

    subgraph LLMGEN["🤖 LLM ANSWER GENERATION"]
        RAG_PROMPT["RAG Prompt Template<br/>Structure:<br/>• System: You are a helpful assistant<br/>• Context: Retrieved chunks<br/>• Question: User query<br/>• Instruction: Answer using context"]:::llm

        LLM_ENGINE["LLM Engine<br/>Model: Qwen 2.5 (14B/32B)<br/>API: Ollama/vLLM<br/>Port: 11434<br/>Temp: 0.2 (factual)<br/>Max tokens: 2048<br/>Stream: enabled"]:::llm

        ANSWER_POST["Answer Post-processor<br/>• Citation formatting<br/>• Source links<br/>• Confidence scoring<br/>• Fallback handling"]:::llm
    end

    CONTEXT_BUILDER -->|"Context<br/>+ Sources"| RAG_PROMPT
    QUERY_PREPROCESS -->|"Original<br/>Question"| RAG_PROMPT
    RAG_PROMPT --> LLM_ENGINE
    LLM_ENGINE --> ANSWER_POST
    ANSWER_POST -->|"Final<br/>Answer"| USER_INPUT

    %% =====================================
    %% LAYER 8: CACHING LAYER
    %% =====================================

    subgraph CACHING["💾 MULTI-LEVEL CACHE"]
        REDIS_CACHE["Redis Cluster<br/>Mode: Cluster<br/>Nodes: 3<br/>Memory: 16GB<br/>Persistence: AOF"]:::cache

        CACHE_TYPES["Cache Types:<br/>• Query embeddings (TTL: 7d)<br/>• Search results (TTL: 1h)<br/>• LLM responses (TTL: 24h)<br/>• Popular queries (no TTL)<br/>Eviction: LRU"]:::cache

        CACHE_WARMING["Cache Warming<br/>Pre-compute:<br/>• Top 100 queries<br/>• Common patterns<br/>Schedule: daily<br/>Update: on doc changes"]:::cache
    end

    REDIS_CACHE --> CACHE_TYPES
    CACHE_TYPES --> CACHE_WARMING

    QUERY_EMBEDDER <-.-> REDIS_CACHE
    HYBRID_SEARCH <-.-> REDIS_CACHE
    LLM_ENGINE <-.-> REDIS_CACHE

    %% =====================================
    %% LAYER 9: SCALING & LOAD BALANCING
    %% =====================================

    subgraph SCALING["📈 SCALING INFRASTRUCTURE"]
        LOAD_BALANCER["Load Balancer<br/>Type: Nginx / HAProxy<br/>Algorithm: Least connections<br/>Health checks: /health<br/>Timeout: 30s"]:::monitor

        QUERY_API["Query API Instances<br/>Replicas: 3-10 (auto-scale)<br/>Lang: FastAPI<br/>Container: Docker<br/>Orchestration: K8s"]:::user

        EMBED_WORKERS["Embedding Workers<br/>Replicas: 4-8<br/>GPU: Optional<br/>Queue: Redis<br/>Auto-scale: based on queue depth"]:::process
    end

    LOAD_BALANCER --> QUERY_API
    QUERY_API --> USER_INPUT

    %% =====================================
    %% LAYER 10: MONITORING & OBSERVABILITY
    %% =====================================

    subgraph MONITORING["📊 MONITORING & ANALYTICS"]
        METRICS["Prometheus Metrics<br/>• Query latency (p50, p95, p99)<br/>• Vector search time<br/>• LLM response time<br/>• Cache hit rate<br/>• Embedding generation rate<br/>Scrape: 15s"]:::monitor

        DASHBOARDS["Grafana Dashboards<br/>• RAG Performance<br/>• Query analytics<br/>• Resource utilization<br/>• Error tracking<br/>Refresh: real-time"]:::monitor

        ANALYTICS["Query Analytics<br/>Track:<br/>• Popular queries<br/>• Failed queries<br/>• Avg relevance scores<br/>• User satisfaction<br/>Storage: TimescaleDB"]:::monitor

        ALERTS["Alerting Rules<br/>• Latency > 5s<br/>• Error rate > 5%<br/>• Cache hit < 70%<br/>• Vector DB down<br/>Channel: Slack + Email"]:::monitor
    end

    METRICS --> DASHBOARDS
    DASHBOARDS --> ANALYTICS
    ANALYTICS --> ALERTS

    QUERY_API -.->|"metrics"| METRICS
    HYBRID_SEARCH -.->|"metrics"| METRICS
    LLM_ENGINE -.->|"metrics"| METRICS
    QDRANT -.->|"metrics"| METRICS

    %% =====================================
    %% LAYER 11: FEEDBACK LOOP
    %% =====================================

    subgraph FEEDBACK["🔄 FEEDBACK & IMPROVEMENT"]
        USER_FEEDBACK["User Feedback<br/>• Thumbs up/down<br/>• Relevance rating<br/>• Comments<br/>Storage: PostgreSQL"]:::user

        FEEDBACK_ANALYSIS["Feedback Analysis<br/>• Identify bad answers<br/>• Track improvement areas<br/>• A/B testing results<br/>Schedule: weekly"]:::monitor

        MODEL_TUNING["Model Fine-tuning<br/>• Re-rank model updates<br/>• Prompt optimization<br/>• Chunk size tuning<br/>Cycle: monthly"]:::process
    end

    USER_INPUT -->|"Rate<br/>Answer"| USER_FEEDBACK
    USER_FEEDBACK --> FEEDBACK_ANALYSIS
    FEEDBACK_ANALYSIS --> MODEL_TUNING
    MODEL_TUNING -.->|"Improve"| RERANKER

    %% =====================================
    %% ANNOTATIONS
    %% =====================================

    SCALE_NOTE["📈 SCALABILITY:<br/>• Vector DB: Horizontal sharding<br/>• API: K8s auto-scaling (HPA)<br/>• Workers: Queue-based scaling<br/>• Cache: Redis cluster<br/>Target: 100k+ docs, 1k+ QPS"]:::monitor

    PERF_NOTE["⚡ PERFORMANCE TARGETS:<br/>• Query latency: <3s (p95)<br/>• Vector search: <100ms<br/>• LLM generation: <2s<br/>• Cache hit rate: >80%<br/>• Throughput: 1000 QPS"]:::cache

    QUALITY_NOTE["✅ QUALITY ASSURANCE:<br/>• Re-ranking for precision<br/>• Source attribution<br/>• Confidence scoring<br/>• Fallback responses<br/>• Human feedback loop"]:::process

    SCALE_NOTE -.-> QDRANT
    PERF_NOTE -.-> REDIS_CACHE
    QUALITY_NOTE -.-> RERANKER

Pipeline RAG

1. Ingestion Pipeline (Offline)

Parsing documentazione MkDocs
Chunking intelligente (512 token, overlap 128)
Generazione embeddings (all-MiniLM-L6-v2)
Storage in Vector Database (Qdrant cluster)

2. Query Pipeline (Real-time)

Embedding della query utente
Hybrid search (semantic + keyword)
Re-ranking con cross-encoder
Context assembly per LLM

3. Generation

LLM locale (Qwen) con RAG context
Source attribution automatica
Streaming delle risposte

4. Scaling Strategy

Vector DB sharding automatico
API instances con auto-scaling K8s
Redis cluster per caching multi-livello
Load balancing con Nginx

📧 Contatti

Team: Infrastructure Documentation Team
Email: infra-docs@company.com
GitLab: https://gitlab.com/company/infra-docs-automation

Versione: 1.0.0 Ultimo aggiornamento: 2025-10-28

32 KiB Raw Blame History

📚 Automated Infrastructure Documentation System

📋 Indice

🎯 Overview

Caratteristiche Principali

🏗️ Architettura

📊 Schema Architetturale

Management View

🔧 Schema Tecnico

Implementation View

💬 Sistema RAG Conversazionale

Interrogazione Documentazione con AI

Caratteristiche Principali

Schema RAG - Management View

Schema RAG - Technical View

Pipeline RAG

📧 Contatti

32 KiB

Raw Blame History