ductrantrong

TRAN TRONG DUC

Zalo/Phone: +84 987 308 065

Email: trantrongducqt@gmail.com

Backend: Python, Flask, FastAPI, threading, multiprocessing
Data Science / ML: PyTorch, TensorRT, ONNX, RKNN, Triton, XGBoost, Optuna, Pandas, Hugging Face
LLM Agents: Google-ADK, mem0, LiteLLM
Computer Vision: OpenCV, Pillow, GStreamer, YOLO, Tracker, PaddleOCR
Databases: MongoDB, SQL, FAISS, Qdrant, Redis, Phoenix
Ops & Dev Tools: Docker, Portainer, MLflow, Pytest, Black, Isort, Flake8, GitLab CI/CD
Cloud: AWS

July 2023 - Present

Fraud Detection System: Designed a Kafka-integrated system processing raw transaction data. Built preprocessing pipelines for feature extraction (using PaddleOCR for images) and developed an ensemble of multi-modal models to predict fraudulent transactions.
Automated Trading Agent: Built an agent utilizing tabular OHLC data with XGBoost and Freqtrade/FreqAI, enhanced by an integrated LLM for real-time news analysis.
Conversational Agents: Developed chat agents powered by Google-ADK with tools for internet search, RAG, database query, and memory management (mem0), using Arize Phoenix for observability.
Search & Recommendation Engine: Implemented an engine with LLM-driven structured extraction, embeddings, and semantic search for personalized recommendations.
Smart Edge Agents: Developed speech-capable agents for low-power boards, utilizing a custom LiteLLM socket middleware for usage management and logging.
MLOps: Built GitLab CI/CD pipelines for automated testing and deployment; integrated MLflow for tracking and Portainer for resource management.

August 2020 - June 2023

Face Recognition System: Designed an end-to-end pipeline with edge capture, Kafka streaming, and server-side recognition (alignment, record management).
Camera Analytics Pipeline: Built a system for customer detection, tracking, counting, and heatmap generation using YOLO, managing the full lifecycle from data labeling to deployment.
Background Removal Pipeline: Developed a U-Net based solution for background removal, covering data collection, training, and deployment.
Edge Applications: Optimized applications on Jetson Nano and Orange Pi using source code encryption and GStreamer for secure transmission.
Model Optimization: Applied quantization, pruning, TensorRT, ONNX, and Triton to enhance inference speed.
Backend Services: Deployed scalable backend services using Flask and Gunicorn.

2016 - 2020