Đặng Vũ Tuấn Kiệt

Implements a real-time WebSocket API for generating lip-synced talking head videos from a single image and live audio input.

Utilizing RabbitMQ for message queuing to facilitate low-latency, real-time video generation and streaming.

Design a system for a smart home with multiple AI features: Facial Recognition, Voice Command Classification, Anomaly Detection.

Collect data from the hardware and use it for training the model.

Develop a small LLM for multiple-choice code question answering.

The model can now incorporate a dynamic number of choices using a learnable token.

Building a deep learning library from scratch just using Python and Numpy.

The library provides a range of features, from multi-layer perceptrons to convolutional neural networks.

Enhance the Round Robin scheduler algorithm based on user requirements using the transformer architecture.

Early results show it can achieve high accuracy while still being fast.

Research and improve the performance of brain tumor segmentation on the BraTS dataset.

The design model achieves impressive performance on BraTS and multiple medical segmentation datasets.

Design a genetic algorithm, a biology-inspired approach, to solve the online puzzle game Reach The Flag.

Design and collect data to create a game environment to visualize and solve the game.