CV View my PDF CV

Contact Information

Name Tran Khanh Thanh
Professional Title Electrical Engineering student and Software Engineer
Email 25thanh.tk@vinuni.edu.vn
Phone +84 364 491 720
Location , Hanoi, Vietnam

Professional Summary

Bachelor of Science student in Electrical Engineering at Vin University with experience building AI, robotics, simulation, and full-stack software systems.

Experience

  • 2025 - Present

    Software Engineer
    STEAM for Vietnam
    Developing software systems for robotics education and competition operations.
    • Built a cross-platform desktop application for competition staff using TypeScript, Bun, Electrobun, React, Hono, and SQLite.
    • Implemented LAN-based local HTTP server capabilities for simultaneous venue operations without internet dependency.
    • Delivered real-time multi-user editing with Server-Sent Events and TinyBase for sub-second synchronization during live match sessions.
    • Designed competition algorithms for scheduling, ranking, rule processing, and practice, qualification, and playoff workflows.

Education

Awards

Skills

Programming Languages (): TypeScript, Python, SQL, C++
Frontend and Desktop Development (): React, Next.js, Three.js, WebGPU, Electrobun, TanStack Start, Zustand
Backend and Infrastructure (): Bun, Hono, FastAPI, PostgreSQL, SQLite, Redis, AWS, Docker, Server-Sent Events
AI, ML, and Research (): PyTorch, LangChain, Manim, Gradio, GiNOT, Retrieval-Augmented Generation, Statistical Analysis

Projects

  • Robotics Tournament Management System

    A cross-platform desktop and local-network application for robotics competition operations, built with Bun, Electrobun, React, Hono, SQLite, TanStack Start, PostgreSQL, Server-Sent Events, and TinyBase.

    • Built a native single-executable desktop application with a 30 MB build size and about 200 MB runtime memory footprint for simple one-click deployment.
    • Implemented a LAN-based local HTTP server so operators on the same network can connect simultaneously without internet dependency.
    • Delivered real-time multi-user editing with Server-Sent Events and TinyBase for optimistic local state and sub-second client synchronization.
    • Designed core competition algorithms for scheduling, dynamic ranking computation, rule processing, and workflows across practice, qualification, and playoff phases.
  • StemFun: AI-powered text to code to video for Education

    An AI-powered system that generates Manim videos from theorem explanations using LangChain, FastAPI, AWS, Redis, Gradio, and Docker.

    • Developed TheoremExplainAgent to generate long-form Manim videos that visually explain mathematical theorems and identify potential reasoning flaws.
    • Designed and implemented a modular architecture for video planning, code generation with optional RAG integration, Manim rendering, and evaluation.
    • Integrated Large Language Models to translate theorem explanations into Manim code and built an evaluation suite to assess video quality.
  • Fast CFD --- 3D Building Editor & ML-Accelerated CFD Simulation for HVAC

    A 3D building editor and ML-accelerated CFD simulation platform for HVAC workflows using Three.js/WebGPU, Next.js 16, FastAPI, PyTorch, GiNOT, and Docker.

    • Developed an interactive drag-and-drop 3D building editor using React 19, Three.js/WebGPU, and Next.js 16 for multi-story HVAC layouts with 8+ component types.
    • Built a FastAPI and PyTorch backend using the GiNOT architecture to replace traditional CFD solvers, achieving about 1,000x faster inference.
    • Implemented STL/OBJ parsing, point sampling, normalization, model caching, and Dockerized REST APIs for end-to-end mesh pipelines.
    • Architected a Turborepo monorepo with 4 packages, a dirty-node ECS rendering pipeline, 50-step undo/redo, CSG Boolean operations, and spatial-grid collision detection.
  • PRISM: A Multi-Dimensional Benchmark for Evaluating LLM Peer Reviewers

    A research benchmark for evaluating AI-generated peer reviews across quality metrics, conference datasets, and human-reviewer comparisons.

    • Developed a multi-dimensional benchmark to evaluate AI-generated peer reviews across 6 quality metrics, comparing 5 LLM baselines against human reviewers over 5 conference datasets.
    • Designed an evidence-grounded evaluation pipeline for novelty assessment, flaw identification, issue prioritization, and constructiveness using retrieval-based verification and statistical validation.
    • Performed Pearson correlation, Wilcoxon signed-rank, and Holm-Bonferroni analyses to show metric independence, quantify model-human gaps, and identify strengths of leading review-generation systems.