A PyTorch-based CNN trained on the Oxford Flower dataset. Increased accuracy from baseline 60% to
85.13% on
the 10-class dataset and 83.63% on the
102-class dataset.
A latent diffusion pipeline including a U-Net denoising model and a variational autoencoder
to
generate flower-like images from pure random noise over 10 iterative denoising steps.
A C++ ray tracing computer graphics rendering engine implementing geometric primitives, Phong
illumination, shadows, mirror
reflections, super-sampling anti-aliasing, and spotlight rendering.
Developed multiple showcase scenes, including a robot, cube/cylinder/quadric/sphere, to
demonstrate
realistic lighting, reflections, and rendering techniques.
Developed an image-based target tracking system using SIFT feature matching, OpenCV, and PnP-RANSAC
for
robust camera pose estimation.
Evaluated multiple correspondence filtering strategies under
challenging conditions including rotation, scale changes, viewpoint variation, and partial occlusion
to improve tracking accuracy and robustness.
OETD: A new EEG (brain signal) dataset with 9200 samples collected from 12 subjects. A comprehensive evaluation across CNN / Transformer / Contrastive Learning / LLM, achieving 27.05% test accuracy on 10-class EEG-to-Text decoding task.
Publication submitted to IEEE ICDM 2026 and under review:
"OETD: Exposing Evaluation Pitfalls in EEG-to-Text Decoding with a Realistic
Benchmark"
Predict wine location based on NIR (near-infrared) and hyperspectral data through the bottle. Machine learning models (Logistic Regression, KNN, SVM, Random Forest, XGBOOST) and CNNs to classify wine types/regions/producers, achieving f1-scores 89.33% on 5-class region and 100% on 3-class liquid type classification.
A keyword search engine featuring an inverted index mapped over a corpus of 173,000 documents. Integrated RocksDB storage utilizing Log-Structured Merge (LSM) trees for optimized write performance and key-value lookups.
An interactive Power BI dashboard/report created on the Adventure Works sales data
(FY2018–2022).
Data Modeling: A Star Schema semantic
model with optimized table relationships.
DAX Measures: Created measures for
time-intelligence, profit margins, and performance variance metrics.
UX & Insights: Drill-through
matrices, conditional formatting, and bookmarks enhance the user interaction experience.
AI-solutions for wine authenticity and traceability: Cleaning, combination and transformation, PCA for near-infrared and hyperspectral wine data; ML/DL model development and evaluation.
GPA: 7.7 / 9
Coursework: Advanced AI, Deep Learning,
Machine Learning & Data Mining, Computer Vision & Graphics, Information Retrieval,
Programming for AI, When Machines Decide
Managed high-value commercial advertising projects for automotive clients on Dongchedi. Cross-functional coordination across media, sales, technology, UI/UX, creative production and platform operations.
