Now: test-time adaptation research on drone imagery
Software engineering @ UIT · Ho Chi Minh City
Thinh Phat Ho — Software Engineering student at UIT, VNU-HCM. I research AI for drone imagery and cloud systems, and build backend systems in C#.
By the numbers
Research · Domain-shift benchmark for drone detection
Test-time adaptation methods are usually scored with one number per dataset, which hides which shift they actually handle. I'm building a benchmark that tags every DOTA and VisDrone image with seven shift attributes, so a method can be judged per shift: does it recover under oblique views but fail in low light? Box and pixel measurements handle what can be counted; Qwen3.8-27B reasons about what needs scene understanding, with a second diagnostic pass for the hardest calls.
Projects & research
2026 · Author · UIT
Adaptive Transformer-based log anomaly detection for reliable cloud and networked systems. Raw logs are parsed with Drain, event sequences are modelled with multi-head self-attention instead of an LSTM, and an entropy-driven Adaptive Top-K threshold keeps detection steady through deployments.
2026 – present · AI researcher · UIT
2026 · Solo · Side project
A pixel pet that lives in Claude Code: it walks, talks, gets thrown and stunned, and cosplays.
2026 · Backend developer · SE361
Backend for an e-learning platform, rebuilt as 7 independently deployable microservices behind a YARP API gateway, with RabbitMQ events between services and Redis caching.
2026 · Team of 3 · SE104
E-learning platform for courses, assignments and timed online exams, with student, teacher and admin roles. A WPF desktop client on an ASP.NET Core Web API, backed by Firestore with JWT auth.
Experience
University of Information Technology, VNU-HCM. GPA 8.77/10 after four semesters.
Led UIT's student ambassadors on visits to local high schools, running logistics and activities for prospective students.
Backend developer on 7 services behind a YARP gateway, with RabbitMQ, Redis and Docker.
Led team UIT-TraDaoCamSa through FTU's data competition: schedule, task split, problem-solving strategy.
First paper: Transformer-based log anomaly detection for cloud and microservice systems.
Building a 7-attribute domain-shift benchmark for drone detection, tagged by Qwen3.8-27B plus computer-vision measurements on an HPC cluster.
I want to learn from a team that ships, and bring what I've learned in research to real products. Open to roles like: