Do Xuan Loc · Engineer & founderPORTFOLIO / 2026
Principal Full-Stack & AI Engineer
AI that works.Systems that last.
I build production AI systems with reliability, security, and operating cost in mind. From the interface to the inference pipeline.
Based in Ho Chi Minh City · Working across borders
Drag to rotate · Select a perspective
A living architecture / concept study
Interface, infrastructure, and the decisions in between.
Models evaluated in the workflow they serve.
Actions that can be inspected. Decisions that can be explained.
01 / Selected work
Complex problems. Clear decisions.
A closer look at how I connect product experience, system architecture, and AI accountability.
Interactive study · demo data
Xemv1.0.3
ConversationRefine the workspace layoutReview the change alongside the project notes.
Repositoryworkspace.ts
const context = {
project: "demo",
panes: 4
}
Diff+ preserveContext()
+ restoreConversation()
Project knowledgedocs / dossier.md
One desk. Multiple contexts.
Keep a conversation, repository, and project notes in view.
demo / workspace
↳Refine the workspace layoutPrevious conversation
↗ Refine the workspace layoutReview the change alongside the project notes.
Continue with context→
Pick up where you left off.
Conversation history travels with the session.
ACTION HISTORYSHA-256
1Read contextcontext.md
a7f…2c9 2Edit workspaceworkspace.ts
b4e…8a1 3Review changesgit diff
c2d…5f0 Follow the action trail.
Inspect recorded actions and their hash-chain history.
Illustrative interface; no live project data.
Developer tools · Valixara · v1.0.301
A macOS workspace for projects, AI sessions, and an inspectable action history. Built for working across repositories without losing context.
The approachBring a multi-pane workbench, resumable conversations, project knowledge, and a hash-chained action ledger into one local-first desktop app.
RustTauri 2React 19SQLite
Explore the work↗ Audio01Inference02Context03
Concept illustration
↗ Production AI · Voice intelligence02
Call recordings, transcription, speaker separation, and structured summaries connected in one workflow.
The approachPair event-driven inference with structured extraction, observability, and evaluation at each stage.
PythonAWSWhisperNeMo
Explore the work↗ Finding01Evidence02Decision03
Concept illustration
↗ Security · Engineering systems03
An AI-assisted security review harness designed around machine-verifiable findings.
The approachCombine established scanners, reproducible checks, and code-assigned verdicts with AI assistance.
PythonLangGraphSemgrepTrivy
Explore the work↗ 02 / The work behind the work
Building. Leading. Learning.
Full-stack foundations, enterprise systems, and a growing focus on production AI and developer tooling.
Explore experience↗ 01Production AI evaluation
Recording quality, transcription, diarization, and speaker roles—evaluated together with the downstream workflow.
02Security verification
Trace findings to evidence, inspect release risks, and make the next action clear for the engineering team.
03AI product operations
Photo and listing-copy workflows, data synchronization, storage lifecycles, and cloud cost.
04Agent tooling & privacy
Local developer workflows, tool permissions, session history, and careful handling of information in logs.
Do Xuan Loc