Back to projects
Production

AI Data Center Monitoring Agent - Client Project

Dashboard platform with embedded bilingual AI chat (EN/ES) for industrial data center monitoring. 4-tier n8n architecture: (1) Config Sync (12h interval) from Radix SCADA API to PostgreSQL via CSV parsing, (2) Realtime Data Pipeline (5min) with batch processing & DGO filtering, (3) Conversational AI Agent with Google Vertex AI processing structured JSON (QUERY/READY/RESPONSE format), contextual memory, and industrial standards analysis (TIA-942, ASHRAE), (4) RAG system with Google Drive ingestion to Vector DB for compliance documentation lookup. Agent generates field-actionable recommendations with urgency levels, timelines, and cost estimates. Monitors generators (Cummins DQKAE), UPS systems, HVAC units, electrical meters. End-to-end delivery: requirements gathering, architecture design, development, deployment, ongoing support.

n8nGoogle Vertex AIPostgreSQLVector DBRAGRadix SCADA APIGoogle EmbeddingsSSE Streaming

Architecture 4-Workflows n8n Orchestrée

Platform with 4 synchronized n8n workflows for industrial monitoring with AI-driven analysis

01

Chat AI Agent Workflow

Conversational pipeline: Webhook → Extract Data → AI Agent (Vertex AI) → Database Query → Parse JSON (QUERY/READY/RESPONSE) → Format Response → Send. Bilingual EN/ES with contextual memory.

n8n WebhookGoogle Vertex AIPostgreSQLSSEConversation Memory
02

Config Sync Pipeline (12h)

Fetch Config from Radix SCADA API → Parse CSV & Filter Internal Points → Generate SQL → Batch Processing → Upsert to PostgreSQL (radix.datapoints_config). Syncs equipment metadata.

Cron TriggerRadix APICSV ParserBatch ProcessingPostgreSQL Upsert
03

Realtime Data Pipeline (5min)

Fetch Realtime from Radix API → Format & Filter → DGO DATA TABLE → Batch Processing → Insert Values to PostgreSQL (radix.datapoints_values). Monitors: generators, UPS, HVAC, meters.

5min CronRadix APIData FilteringBatch InsertsPostgreSQL
04

RAG System Workflow

RAG AI Agent with Chat Interface + Agent Tools for RAG (embeddings Google Vertex) + Google Drive File Ingestion to Vector DB. Compliance documentation lookup (TIA-942, ASHRAE).

RAG AgentVector DBGoogle EmbeddingsGoogle DriveDocument Indexing

n8n Workflows Architecture

Workflow 1: Chat AI Agent

Conversational Pipeline with Google Vertex AI

Webhook → Extract Data → AI Agent → Database Query → Parse JSON → Format Response → Send

Chat AI Agent Workflow
Workflow 2: Config Sync (12h)

Equipment Metadata Synchronization

Radix API → Parse CSV → Generate SQL → Batch Upsert

Config Sync Workflow
Workflow 3: Realtime Data (5min)

Sensor Telemetry Ingestion

Fetch Realtime → Filter → Batch Insert to PostgreSQL

Realtime Data Workflow
Workflow 4: RAG System

Compliance Documentation Retrieval

RAG Agent + Google Drive Ingestion → Vector DB → Embeddings for standards lookup

RAG System Workflow

Key Features

Bilingual AI Chat (EN/ES)

Conversational agent with Google Vertex AI. Contextual memory for multi-turn conversations. Natural language queries on industrial equipment

Industrial Standards Analysis

Real-time compliance checking against TIA-942 (datacenter) and ASHRAE (HVAC) standards. Automated temperature, humidity, power analysis

Field-Actionable Recommendations

AI generates specific on-site actions with urgency levels (Immediate/Urgent/Scheduled/Routine), timelines, required personnel, and cost estimates

Structured JSON Parsing

Strict JSON format (QUERY/READY/RESPONSE) enforced via system prompt. Query generation for PostgreSQL, status flags for workflow orchestration

Dual-Pipeline Data Ingestion

Config sync (12h) for equipment metadata + Realtime data (5min) for sensor values. Radix SCADA API integration with CSV parsing & batch processing

RAG Documentation Lookup

Vector DB with Google embeddings for compliance docs. Google Drive file ingestion. Agent retrieves relevant standards during analysis

Multi-Equipment Monitoring

Generators (Cummins DQKAE: 15 metrics), UPS (Eaton: 7 metrics), HVAC (Daikin: 6 metrics), electrical meters. 28+ datapoints per site

Get Real Time API Tool

HTTP tool integrated with AI agent for fetching latest single-metric values. XID-based queries to Radix SCADA API for real-time data fresher than database

Tech Stack

n8n Workflow Engine Orchestration

4 workflows (Chat Agent, Config Sync 12h, Realtime 5min, RAG). 20+ nodes per workflow with cron scheduling

Google Vertex AI LLM

Conversational AI with structured JSON output (QUERY/READY/RESPONSE). Bilingual EN/ES support

PostgreSQL Database

radix.datapoints_config (equipment metadata) + radix.datapoints_values (time-series). Indexed by xid, device_id, timestamp

Radix SCADA API Data Source

Config endpoint (CSV export) + Realtime endpoint (latest sensor values). OAuth/JWT authentication

Vector DB RAG

Google embeddings for compliance documentation. Semantic search for TIA-942, ASHRAE, manufacturer specs

Google Embeddings ML

text-embedding-004 for document vectorization. RAG retrieval during agent conversations

Conversation Memory AI

PostgreSQL-backed session storage. Cross-turn context retention for multi-step troubleshooting

Get Real Time Tool API Integration

HTTP tool for AI agent. XID-based queries to Radix API for fresh single-metric values

CSV Parser ETL

Config sync pipeline. Header validation, type coercion, internal point filtering

Batch Processing Performance

Both pipelines use batch inserts/upserts (100-500 rows). Reduces DB round-trips, optimizes throughput

Results & Metrics

Technical Performance

3-8 sec
Chat Response Time
Webhook → Extract → AI → Query → Parse → Format
~2 min
Config Sync Duration
28 datapoints, CSV parse, batch upsert
< 5 min
Realtime Data Latency
5min cron, batch processing, PostgreSQL insert
Cross-session
Conversation Memory
PostgreSQL-backed context retention

Business Impact

4 types
Equipment Monitored
Generators, UPS, HVAC, Meters
28+
Datapoints per Site
SITE_TRIARA coverage
2 frameworks
Standards Compliance
TIA-942 + ASHRAE automated checks
2
Languages Supported
English + Spanish bilingual

Security & Compliance

Prevented
SQL Injection
Parameterized queries, no dynamic SQL
~98%
JSON Format Compliance
Strict QUERY/READY/RESPONSE enforcement
Site-based
Data Isolation
PostgreSQL schema radix.* separation
OAuth/JWT
API Authentication
Radix SCADA API secured access

Technical Challenges & Solutions

Enforcing Strict JSON Format from LLM

Problem
LLMs tend to output markdown, explanations, or malformed JSON. System needed strict QUERY/READY/RESPONSE structure for workflow orchestration
Solution
System prompt with CRITICAL section enforcing JSON-only output. Example-driven prompt with forbidden/required patterns. Parse & Decide node validates JSON before processing

Multi-Source Data Synchronization

Problem
Config (12h) and Realtime (5min) pipelines must stay consistent. Config changes could orphan realtime data. Radix API returns CSV with inconsistent schemas
Solution
Config pipeline upserts with conflict resolution (ON CONFLICT DO UPDATE). Realtime pipeline references datapoints_config.xid. Foreign key constraints prevent orphans. CSV parsing with header validation

Industrial Standards Compliance Logic

Problem
AI must check TIA-942 (18-27°C, 40-60% RH) and ASHRAE ranges, Cummins generator specs (70-90°C coolant, 1500-1800 RPM), UPS battery limits (20-25°C optimal)
Solution
System prompt embeds full standards reference (100+ lines). datapoints_config.min_value/max_value provide equipment-specific bounds. AI cross-references both during analysis

Real-Time vs Historical Data Reconciliation

Problem
"Get Real Time" tool fetches fresher data than DB but requires XID lookup. User asks "current temp" but XID unknown. Multiple XIDs match query
Solution
2-step workflow: (1) Query datapoints_config for XID via ILIKE name search, (2) Use XID with Get Real Time tool. If multiple XIDs, prompt user for clarification

Field Action Recommendations with Cost/Timeline

Problem
Generic responses ("check the system") not actionable. Need specific tasks, urgency (same day vs 2 weeks), personnel (technician vs specialist), estimated costs
Solution
System prompt templates for 4 urgency levels (Immediate/Urgent/Scheduled/Routine) with required fields: timeline, personnel, parts/costs, risk if ignored. AI trained on real maintenance scenarios

RAG Document Retrieval for Compliance

Problem
Agent needs to reference TIA-942 Annex F, ASHRAE TC 9.9 guidelines, Cummins service manuals during conversations. Docs not in prompt due to token limits
Solution
Separate RAG workflow: Google Drive ingestion → Extract Document Text → Embeddings Google Vertex → Vector DB indexing. Chat agent queries Vector DB for relevant doc chunks via semantic search

Demonstrated Skills

AI/LLM Engineering

Structured JSON EnforcementPrompt Engineering (2000+ token system prompt)RAG ArchitectureTool Use (Function Calling)Bilingual Agent (EN/ES)

Workflow Orchestration

n8n Multi-Workflow DesignCron Scheduling (12h/5min)Webhook IntegrationError Handling & Retry LogicParallel Pipeline Execution

Data Engineering

ETL Pipeline DesignCSV Parsing & ValidationBatch Processing (100-500 rows)Time-Series Data ModelingPostgreSQL Schema Design

Industrial Monitoring

SCADA API IntegrationEquipment Telemetry (Generators/UPS/HVAC)Standards Compliance (TIA-942/ASHRAE)Field Action PlanningReal-Time Alerting

Client Delivery

Requirements GatheringArchitecture DesignEnd-to-End DevelopmentDeployment & TestingOngoing Support & Maintenance

Interested in this project?

Contact me to discuss similar projects or for more information.

Contact me View all projects