I build big data platforms — from infrastructure to semantic layers — and have spent the last few years exploring how LLM-powered multi-agent systems can make analytics faster and more honest. Mostly working with AWS, Snowflake, Databricks, and Python.
Twelve flagship projects across five companies — dive into a focused write-up per project under Recent Work or read the long-form story in Posts.
mindmap
root((12 Flagship Projects))
FPT Canada
P1 Click-stream Web Analytics
200B+ Omni-Channel Ingestion
10M-Entity MDS
Self-Serve Data Platform
Omni-Channel Dashboard Rebuild
P2 Revenue-Leakage Reporting
Receivables Payables Reconciliation
Leakage Detection Alerting
Executive Reporting Audit Trail
P3 MADA Multi-Agent Analytics
Self-Serve LLM Reporting
Always-On Data Quality
Self-Learning Rules Memory
Harness Engineering
Tyme GOTyme Bank
P1 Banking Business Products
Personal Lending
VAS
GoalSave Kicker
ID Payment
P2 Infrastructure
Cert Pinning 1M users
Kiosk Network Optimization
AWS Infra POC
P3 Analytics
500M Tx Reporting Pipeline
Multi-Source Analytics Tooling
NFQ Asia
P1 SEO Keyword Analytics
NLP at 30TB
Serverless 50K req day
Datadog Cost Audit
P2 PoS Serverless API
PoS Device API Layer
Event-Driven Pipeline
Auto-Scaling Cost Control
NashTech Global
P1 Banking Insurance KYC
P2 Process and People
FPT Software
P1 DirecTV Multi-Vendor Delivery
P2 Cebu Dev Center and Training
Selected projects from the last few years. Each card opens a deep-link write-up — share the URL and it lands directly on the case study.
6-agent LangGraph pipeline (Orchestrator, Ingest Auditor, Anomaly Detector, Evidence Retriever, Evaluator) auditing $1.5B in financial transactions. Days → minutes time-to-insight, with citations, retry loops, and a self-learning rule memory.
Bronze→Silver→Gold medallion on AWS + Snowflake + Spark, 99.97% enrichment quality, query latency 30s → 8s, used by 7+ DS / Analytics / Finance teams.
Lambda + Snowflake + DynamoDB pipeline; SES routing to owners; promoted to Data Integrity Lead. End-of-month → next-day; detection 30% → 99.99%.
PKI + mobile SDK + zero-downtime rotation pipeline protecting 1M+ mobile-banking users against MitM. Standardized integration patterns across 5 dev teams.
TF-IDF + lemmatization + POS tagging pipeline; serverless API serving 20–50K req/day; latency 60s → 3s; resolved a $10K/mo Datadog cost issue.
Serverless résumé + portfolio: CloudFront + S3 + Lambda + DynamoDB, Terraform-managed. Includes a RAG chat widget, admin analytics, and a posts CMS.
Long-form notes on what I'm building, learning, and shipping. Filter by category below.
From training engineers in Vietnam & the Philippines to leading AI data engineering in Canada.
timeline
title 17+ Years in Engineering Leadership
2009 : University Lecturer (DNTU / NLU)
2014 : FPT Software - Training Manager / Solution Architect (PH and VN)
2016 : NashTech Global - Principal Engineer / Architect
2017 : NFQ Asia - Technical Architect (NLP at 30TB)
2019 : Tyme / GOTyme Bank - Architect Community Lead (1M+ users)
2021 : FPT Canada - Data Engineering Lead (200B+ events, LangGraph agents)
Six pillars across AI, data, cloud, and security.
mindmap
root((Vinh X. Nguyen))
AI Agentic LLM
LangGraph / LangChain
MCP
RAG and Vector DBs
Anomaly Detection
Data Modeling
OLTP / OLAP 200B+
Star / SCD
Data Vault 2.0
Medallion
Data Engineering
Spark / Databricks
Kafka / Kinesis
Snowflake / Delta Lake
Lambda / Sagemaker
Cloud
AWS Full-Stack
Snowflake
Databricks
Terraform / CFN
Analytics
Tableau / PowerBI
Datadog / Grafana
NLP TF-IDF
Security
PKI / Cert Pinning
Cryptography
WAF / IAM
Event-driven
FPT Canada — LLM agents auditing $1.5B in financial transactions.
flowchart LR
Tx[(Transactions DW)] --> Orchestrator{LangGraph Orchestrator}
Orchestrator --> Rules[Rules Agent]
Orchestrator --> Stats[Statistical Agent]
Orchestrator --> LLM[LLM Reasoning Agent]
LLM --> RAG[(Policy RAG / Vector DB)]
LLM --> MCP[MCP Tools to DBs]
Rules --> Reviewer[Reviewer Agent]
Stats --> Reviewer
LLM --> Reviewer
Reviewer --> Human[Human-in-the-Loop]
Reviewer --> Cases[(Case Store)]
FPT Canada — 99.97% enrichment quality across AWS, Snowflake, Databricks.
flowchart LR
Sources[(Sources)] --> Stream[Kinesis / Kafka]
Stream --> Bronze[S3 Bronze]
Bronze --> Spark[Spark on EMR / Databricks]
Spark --> Silver[S3 Silver / Delta]
Silver --> Gold[S3 Gold / Delta]
Gold --> SF[(Snowflake DW)]
Gold --> Glue[Glue Catalog]
SF --> BI[Tableau / PowerBI / QuickSight]
Spark --> QA[Quality and Reconciliation]
QA --> Alerts[CloudWatch / Datadog]
GOTyme Digital Bank — mitigating MitM at mobile-banking scale.
flowchart LR
CA[Internal CA / PKI] --> Cert[Server Cert + Backup Pin]
Cert --> SDK[Mobile SDK]
SDK --> App[Banking App 1M+ users]
App --> API[(Banking APIs)]
Rotation[Rotation Pipeline] --> Cert
App --> Telem[Telemetry / Crash Analytics]
Telem --> Flags[Feature Flags / Staged Rollout]
NFQ Asia — in-store PoS hardware integrated with auto-scaling AWS backend.
flowchart LR
PoS[PoS Devices] --> APIGW[API Gateway]
APIGW --> Lambda[AWS Lambda Edge API]
Lambda --> SQS[SQS Queue]
SQS --> Worker[Lambda Worker]
Worker --> DDB[(DynamoDB)]
Worker --> EB[EventBridge]
EB --> DWH[(Warehouse / Analytics)]
Lambda --> CW[CloudWatch / Alarms]
AI-generated answers grounded in Vinh's resume & this site. May contain errors.