VAICENTRAL: Vibe Apps AI Central Platform
VAICentral is a next-generation Unified AI & Data Platform that eliminates data silos by bringing structured data (ERP/DWH) and unstructured data (documents, SOPs, procedures) into a single architecture—powered by GraphRAG and a Local LLM Engine.
Technical overview
Four layers from business users through AI gateway, unified AI + data warehouse platform, to on-premise infrastructure and observability.
Reference blueprint: NL query interface, AI agents, GraphRAG core, document pools, ClickHouse analytics, and enterprise governance on one stack.
Core modules & features
Four subsystems form the VAICentral brain—orchestrating AI, digitizing knowledge, governing analytics, and enforcing enterprise guardrails.
AI Orchestration & GraphRAG Core
The brain of VAICentral: understands user intent and coordinates AI agents to answer across unstructured documents and DWH metrics.
Deep Vietnamese Natural Language Query (NLQ)
Converts everyday Vietnamese questions into database commands—SQL for ClickHouse or Cypher for Neo4j—to fetch real-time data.
Hybrid Search
Combines vector search via Qdrant (similar passages) with graph search via Neo4j (logical relationships between entities).
Multi-hop Reasoning
Links scattered data to answer chain-style questions—for example: procedure document → production batch → distributor → sales report.
Enterprise Content Management & Doc Processor
Ingests, processes, extracts, and digitizes your full document knowledge base (PDF, Word, Excel, images) into Pool 1 (Raw Storage).
On-Premise OCR Engine
Uses PaddleOCR / Tesseract to recognize text from scanned documents, invoices, and acceptance records—without sending data to external cloud APIs.
Smart Chunking & Auto-Tagging
LlamaIndex/LangChain splits documents by logical structure (preserving tables and diagrams) and LLM auto-classifies tags such as department, document type, and related entities.
Version Control & Deduplication
Hash pipeline detects duplicate raw files. When an SOP or procedure is updated, the system overwrites, keeps audit history, and refreshes vector weights.
Unified DWH Analytics & Reporting Lineage
Manages structured data flows from core systems (SAP, CRM, core DB) into Pool 2 (Analytics Storage) for dynamic management reporting.
Automated ELT/CDC Pipeline
Airbyte and Apache Kafka capture change events in real time or hourly batches from ERP into ClickHouse, orchestrated centrally via Apache Airflow.
DataMart Modeling with dbt
Standardizes raw data into business DataMarts (Finance, Revenue, HR) and auto-draws lineage from source tables to final charts.
Secure Embedded BI
Embeds Apache Superset dashboards in VAICentral via Embedded SDK with Guest Token (JWT) and row-level security (RLS) for data access.
AI Governance & Guardrails
Keeps enterprise AI safe and accurate—reducing hallucination and controlling infrastructure cost.
LLM Tracing & Cost Monitoring (Langfuse)
Logs the full path of each question, measures latency, context accuracy, and token consumption.
Approved Process Workflow
Review workflow for inbound knowledge—only documents marked Approved by management are loaded into Vector/Graph DB for AI learning.
Human-In-The-Loop (HITL)
For sensitive, legal, or financial answers, AI drafts are held for specialist review and edit before release to partners or reporting systems.
Productivity signal
Up to ~80% efficiency
Analysts and department staff stop spending hours hunting Excel files, old procedures, or manual data cleanup—AI prepares clean, ready-to-use data.
Strategic business benefits
VAICentral gives enterprises one governed entry point for operational knowledge and business numbers.
Break data silos completely: one gateway for business metrics (SAP/CRM tables) and operational knowledge (PDF procedures and handbooks).
Optimize up to 80% of analyst and staff productivity—no more chasing files or manual data prep; AI acts as a data-ready assistant.
Real-time leadership decisions: the board sees live dashboards or asks in Vietnamese for revenue and cash flow instead of waiting 2–3 days for consolidated reports.
Absolute security and data sovereignty: 100% enterprise open-source stack, on-premise deployment—confidential data never leaves your firewall.
Industrial use cases
Proven scenarios across regulated and operations-heavy industries.
Pharmaceuticals
Pharmaceutical industry focus
Batch/Lot traceability & complaint handling
Enter a lot code and VAICentral maps production date, raw-material suppliers, equipment, QC test PDFs, and distributors/hospitals sold to—from Neo4j graph and ClickHouse sales data.
Regulatory & SOP compliance lookup
Ask how to clean a capsule machine when switching from Paracetamol to Ibuprofen—AI extracts exact SOP steps, required forms, and cites page and PDF source.
Banking & Finance
Credit policy & fraud detection
Fraud detection via graph
Kafka CDC streams card and transfer events; Neo4j maps hidden links. AI alerts when accounts share phone numbers, IPs, or circular money flows.
Credit policy & legal document assistant
Ask maximum overdraft for seafood exporters with inventory collateral—AI synthesizes internal circulars and cross-checks current limits on the DWH.
Manufacturing & Supply Chain
BOM analytics & maintenance knowledge
BOM analytics & cost simulation
If steel prices rise 15%, AI pulls BOM tree from Graph DB and runs SQL on ClickHouse invoice history to forecast product cost and gross margin impact.
Predictive maintenance knowledge
Enter robot error code ERR_404—AI reads vendor PDFs for cause and repair steps and checks ClickHouse for spare parts still in stock.