Skip to content
Ho Dinh Tri

About

Hi, I’m Tri.

AI Security & Software Engineer

I’m Tri — an Information Security student at VNU-HCM University of Science (GPA 3.84/4.0). My focus is AI security: the place where machine learning, software and infrastructure have to be trustworthy at the same time.

Most of my work starts with one question: what breaks if this model, policy or pipeline is wrong? In clinical de-identification, a missed name is a privacy leak. In infrastructure scanning, a flood of false positives teaches engineers to ignore the scanner. The answer usually shapes the architecture more than the model choice does — deterministic validation gates, leakage-safe evaluation, access control enforced at the data layer.

What I bring#

  • AI engineering — LLM data pipelines (1M documents, 9 LLM providers, 13 validation gates), evaluation and retrieval with hybrid search.
  • Security — IaC and policy-as-code research, OWASP API controls, PII de-identification, SARIF and SBOM reporting.
  • DevOps & cloud — Docker, Kubernetes, Terraform, GitHub Actions and CI security gates; Azure and AWS basics.
  • Software engineering — FastAPI and PostgreSQL backends, Next.js, and systems programming in C.

I try to report results the way I would want to read them: with the baseline, the evaluation setup, and what the number does not prove.

What I’m looking for#

An internship in AI engineering, security (application, cloud or AI security), DevOps / platform engineering or backend software engineering — part-time (3–4 days a week) now, and full-time from September 2027.

Experience

  1. May 2026 – Present

    AI Engineer (Part-time) — Multilingual Clinical De-identification

    Meddies · Ho Chi Minh City

    • Co-first author of Meddies-PII (arXiv:2609.12544), a multilingual PII extraction framework for clinical de-identification across 17 languages and 9 PII labels.
    • Design and maintain the attribute-conditioned synthetic data pipeline (9-LLM generator pool) that produced 1M documents filtered by 13 deterministic validation gates.
    • The data trains a 350M-parameter BIOES token classifier scoring 0.827 mean exact-match micro-F1 on 15 external benchmarks (strongest baseline 0.658).
    • Building meddies-faker, which samples personal attributes from country-specific real-world distributions for 20 locales.

  2. Feb 2026 – Present

    Undergraduate Researcher — Infrastructure-as-Code Security

    IaC Lab, Faculty of IT, HCMUS · Ho Chi Minh City

    • Evaluated Checkov across 29 runners and 1,346 CKV rules, studying false-positive reduction through AST context propagation.
    • Evaluated DevSecOps reporting (SARIF, CycloneDX/SPDX SBOM) and CI gating with Bandit and secret scanners across 3,700+ test suites.
    • Curating an open IaC security dataset and classification taxonomy as a reusable research benchmark.

Education

  1. Sep 2024 – Expected Jun 2028

    B.Sc. in Information Technology (Information Security)

    University of Science, VNU-HCM · Ho Chi Minh City

    • GPA 3.84 / 4.0.
    • All coursework completed by mid-2027; final year reserved for a full-time internship.

Recognition

Publications

Awards

  • RMIT Hackathon 2026 — 6th overall

    RMIT University · 3rd at Saigon South campus · 4th in Vietnam · team of 4

    2026

  • Academic Merit Scholarship — awarded twice

    Faculty of IT, University of Science, VNU-HCM · Semester 1 of 2024–2025 and Semester 1 of 2025–2026

    2025 & 2026

Certifications

Skills

AI / ML
PyTorchHugging Face TransformersToken classification (NER, BIOES)LoRA / QLoRARAGEmbeddingsXGBoost
Data & evaluation
Synthetic data generationLLM data pipelinesData validationEntity-level F1 benchmarking
Security
CheckovtfsecBanditSecret scanningOWASP Top 10 / API SecurityRBAC / ABACJWTSBOM (CycloneDX / SPDX)SARIFPII de-identification
Cloud & DevOps
DockerKubernetesTerraformGitHub ActionsAzureAWS (EC2, S3)CaddyLinux
Programming
PythonC / C++SQLBashTypeScriptFastAPIPostgreSQL / pgvector
Systems
Linux / POSIXFile systems & journalingConcurrencyRISC-V (xv6)GDB
Languages
Vietnamese (native)English — C1 (Oxford Placement Test)