Huu Huy Truong
Technology Leader & Enterprise, Data & AI Architect
Enterprise, data and AI architect with over 20 years of experience across public-sector, financial-services and consulting environments. Designs cloud-native data and integration platforms, from enterprise analytics, data lakes, data warehousing, Master Data Management and governance to Machine Learning and Deep Learning experimentation and Agentic AI integration. Combines Azure and AWS data technologies — including Azure Synapse, Databricks, Apache Spark, Azure Data Factory, dbt, Airflow and Kafka — with enterprise integration through MuleSoft, APIs and event-driven architectures. Recent applied AI initiatives include an MCP-enabled Agentic AI Service Hub and a knowledge-base-connected enterprise chatbot.
Public career, expertise and project pages follow the resume of 7 October 2026. The street address is not published, and the resume file is not offered for download.
- 01CareerRoles from 2003 to the current integration architecture lead post, taken from the resume.
- 02Architecture LabEducational reference architectures. Illustrative studies, not client deployments.
- 03ExpertiseData, integration, cloud and applied AI topics from the resume. Proficiency is stated only where the resume states it.
- 04ProjectsSelected engagements from the resume, including WePulse, applied AI work and enterprise analytics platforms.
- 05InsightsArticles on enterprise architecture, data platforms and applied AI.
- 06Contact & ResumeEmail, phone and LinkedIn. The street address and a resume download stay unpublished.
02 · Selected studies
Architecture Lab
A short selection from the lab. Each study walks through one reference architecture: how its parts connect, the design decisions and the trade-offs.
Reference architecture - illustrative, not a client deployment
01 · LLM & Agentic AI
How Enterprise RAG Actually Works
Follow enterprise documents from permission-aware indexing to evidence-based answers, and see why abstention belongs in the design.
Intermediate
02 · LLM & Agentic AI
From Prompt to Action: Anatomy of an AI Agent
See how a proposed tool call becomes an authorized business action through deterministic checks, approval and validated feedback.
Intermediate
03 · LLM & Agentic AI
Multi-Agent Systems Without the Hype
See when a coordinator and scoped specialists earn their cost, and when one bounded agent should write the result.
Intermediate
04 · LLM & Agentic AI
An AI Copilot Inside a Business Process
Follow a support ticket from an authorized draft to a human-approved update, and stop the write when the proposal is rejected or invalid.
Intermediate
Projects
WePulse integration and data-flow architecture
Architecture lead for integration, data flows and shared platform services within WePulse, a €70M multi-tenant SaaS transformation programme for the municipalities of the Brussels-Capital Region.
Agentic AI Service Hub
Designed and prototyped an Agentic AI Service Hub that uses MuleSoft as the enterprise orchestration layer and Model Context Protocol connectors to expose internal services and tools to AI agents.
Knowledge-base enterprise chatbot
Developed a knowledge-base-connected enterprise chatbot, integrated with internal information sources through reusable connectors.
City of Brussels data and digital platforms
Led cloud, data and application architecture for digital-transformation programmes, including SmartBuilding initiatives, for the City of Brussels and its public institutions.
Insights
Architecture Decisions: Balancing Cost, Latency and Control
Cost, latency and control pull against each other in every AI system. A useful decision record names the one a use case favors, what it gives up, and the measurement that would reopen the choice.
Where Deterministic Workflows Should Bound AI Agents
An agent earns its place where the path cannot be written down in advance. Authorization, limits, validation and approvals belong in deterministic code around it, so the model works inside a boundary it cannot change.
Why AI Architecture Starts with Evaluation
A component built on a language model has no testable specification until its evaluation exists. Writing the cases and the acceptance bar first turns model, retrieval and prompt choices into measured architecture decisions.