Help & Supportcareers@gravityer.com
Senior Solutions Architect at NVIDIA
AI is analyzing your overall score…
Identifying your key strengths…
Evaluating your skill match against the job requirements…
Assessing your cultural and operational fit
Our Eyes Are Yet To Open https://github.com/javelin1992?tab=repositories
UC Irvine
Master of Science (M.Sc.), Statistics
January 1, 2015 – January 1, 2017
University of Liverpool
Bachelor of Science - BS, Mathematics with Finance
January 1, 2011 – January 1, 2015
NVIDIA
Senior Deep Learning Solutions Architect
March 1, 2020 – Present
Beijing, China
SUPERG.AI
Computer Vision Algorithm Engineer
November 1, 2018 – March 1, 2020
Beijing, China
Fellowship.AI
Machine Learning Fellow
January 1, 2018 – May 1, 2018
San Francisco Bay Area
Industrial and Commercial Bank of China
Intern
June 1, 2014 – September 1, 2014
Changzhou-Wuxi-Suzhou Metropolitan Area
Sequence To Sequence Language Translator(Udacity)
August 1, 2017 – Present
•Built a machine translator to translate English to French using tensorflow. •Constructed the translator under Encoder-Decoder, Sequence-to-Sequence framework, with stacked 256 units LSTM cell. •Trained and validated the model using a subset of WMT10 French-English corpus dataset, with 137861 sentence pairs, and achieved 0.9904 train accuracy, and 0.9766 validation accuracy.
Book Recommendation System With Implicit Feedback
July 1, 2017 – Present
•Built a recommendation system for books using Book-Crossing dataset. •Constructed user-book interaction matrix, which contained 11,116,762,325 user-book interactions, and user feature matrix, which contained one-hot encoded user features, using pandas, scipy, sklearn, and numpy. •Trained recommendation system using LightFM library by minimizing WARP loss, and achieved 0.94 training AUC score and 0.63 test AUC score.
Limited Training Set Image Recognition Using Transfer Learning
June 1, 2017 – Present
•Classifying classic STL-10 dataset with pre-trained Inception-V3 Network in tensorflow. •Replaced the final softmax classifier of the original Inception-V3 network with a new softmax layer, designed to classify 10 classes instead of 1000, and trained the final layer with 5000 training images, while keeping other trained weights frozen. •Achieved a 96.5% top 1 correction rate on a separate test set with 8000 images.
Character Level Recurrent Neural Network Text Generator
May 1, 2017 – June 1, 2017
• Trained a character level RNN text generator in tensorflow. • Built a RNN with an unrolled length of 150, using stacked 3-layer LSTM cell with 512 neurons, and trained the network on a collection of works by William Shakespeare, which has a total character length of 4,573,338. • Generated new quasi-Shakespeare text with various character length.
Deep Dream Image Generation
May 1, 2017 – June 1, 2017
• Implemented Deep Dream Algorithm based on a pre-trained 22-layer GoogLeNet trained on ImageNet data set, using tensorflow. • Experimented various ways, such as taking a subset of feature maps and using guide images, to effectively regularize the dreaming process in order to reduce cluttering of undesired features. • Generated new images by combining the content and style of two images through guided deep dream.
Credit Card Fraud Detection
March 1, 2017 – Present
•Classifying credit card transactions into either genuine or fraudulent using a highly unbalanced dataset, which contains 490 fraudulent transactions out of 284,807 total transactions. •Trained a benchmark model, which achieved 0.95 AUC score on the test set using LightGBM. •Trained and Tuned a 3-hidden-layer neural networks with dropout regularization by minimizing weighted cross-entropy loss using Tensorflow, and achieved 0.975 AUC score on the test set.
Chaotic Hamiltonian Monte Carlo Sampling
February 1, 2017 – March 1, 2017
•Implemented using python an adaptation of traditional HMC algorithm, which exploits the freedom in choosing momentum distribution to create chaotic sampling trajectories, on simulated 30-dimension Gaussian with various covariance structures. • Compared the sampling results of CHMC to traditional HMC, CHMC seemed to produce samples with negligible correlations, and its performance also seemed to be significantly less reliant on the choice of step-size during numerical integration.
Power and Sample Size Analysis For a Prospective Clinical Trial
November 1, 2016 – January 1, 2017
• Investigated the relationship between sample size and statistical power for a design stage clinical trial to provide insights into how many patients should be included in the final trial. • Modeled the bootstrap simulated data with either generalized linear model, or mixed effects model depending on the level of similarity between involved medical units. • Visualized the Monte Carlo estimated relationship between power and sample size with ggplot2.the data.
Rainfall Prediction With boosted decision Trees and Neural Networks
November 1, 2016 – December 1, 2016
• Predicted the probability of rainfall at a location, based on (processed) infrared satellite image information. • Trained predictive Models based on boosting(Adaboost, Gradient-boost) of basic decision trees, using Python, and regularized model with bagging and L-2 regularization. • Tuned model parameters through 3-fold cross validation and grid search, and final model obtained a 0.8 AUC score on a separate test set.
Generalized Linear and Mixed Effects Models For The Effects Of D-CARB In Reducing Polyp Development
June 1, 2016 – Present
•Measured the effects of coupling D-CARB, a cancer cell inhibitor, and Aspirin had upon preventing cancer cells' growth, and the potential risk of hearing loss after the treatment. • Modeled effects of the coupling treatment on the rate of developing certain cancer cells with a Poisson regression, using R. • Measured risk of hearing loss under a linear mixed effects model, using R, based on the data of a longitudinal follow-up study.
Classification of Higgs Boson Tau-Tau decays with bagged dropout neural network
February 1, 2016 – March 1, 2016
• Distinguished the signals of Higgs-Boson to Tau-Tau decays from severe background noise based on data from the Atlas experiment. • Performed dimensionality reduction based on principal component analysis to extract most prominent signals • Trained bootstrap aggregated 3-hidden-layer feed-forward neural networks with dropout regularization on training set, using R's H2O library, and the optimal model chosen, based on a 3-fold cross-validation, obtained about 85 percent classification correction rate on a separate test set.
Intro to Python for Data Science
DataCamp
June 24, 2026 – Present
Statistical Inference
Coursera Course Certificates
June 24, 2026 – Present
R Programming
Coursera Course Certificates
June 24, 2026 – Present
Exploratory Data Analysis
Coursera Course Certificates
June 24, 2026 – Present
Deep Learning In Python
Udemy
June 24, 2026 – Present
The complete SQL Bootcomp
Udemy
June 24, 2026 – Present
Deep Learning Foundation
Udacity
June 24, 2026 – Present
Introduction to Data Visualization with Python
DataCamp
June 24, 2026 – Present
Neural Networks and Deep Learning
Coursera Course Certificates
June 24, 2026 – Present
Improving Deep Neural Networks: Hyperparameter tuning, Regularization and Optimization
Coursera Course Certificates
June 24, 2026 – Present
Structuring Machine Learning Projects
Coursera Course Certificates
June 24, 2026 – Present
Supervised Learning with scikit-learn
DataCamp
June 24, 2026 – Present
Interactive Data Visualization with Bokeh
DataCamp
June 24, 2026 – Present
Intermediate Python for Data Science
DataCamp
June 24, 2026 – Present
Cleaning Data in Python
DataCamp
June 24, 2026 – Present
APIS AND WEB SCRAPING
Dataquest.io
June 24, 2026 – Present
SQL AND DATABASES: INTERMEDIATE
Dataquest.io
June 24, 2026 – Present
SQL AND DATABASES: ADVANCED
Dataquest.io
June 24, 2026 – Present
Convolutional Neural Networks
Coursera
June 24, 2026 – Present
Sequence Models
Coursera
June 24, 2026 – Present
Deep Learning Specialization
Coursera
June 24, 2026 – Present
Computer Vision Nanodegree
Udacity
June 24, 2026 – Present
Cultural Fit Analysis
The candidate's project portfolio is heavily skewed towards advanced machine learning and deep learning, which aligns well with a data-driven culture. The personal projects demonstrate initiative and a passion for the field. However, the target role is 'Data Analyst,' which often requires a broader skill set in business intelligence, reporting, and stakeholder communication, beyond pure model development. While the candidate has strong analytical skills, the direct alignment with typical Data Analyst responsibilities (e.g., dashboarding, SQL for business insights, A/B testing design) is not explicitly highlighted in their experience or projects. The 'boring data analysis works' description for an internship is a minor red flag regarding enthusiasm for foundational data analysis tasks.
Soft Skills & Operational Fit
The candidate's project descriptions indicate a strong problem-solving aptitude and a methodical approach to data analysis and model development. The variety of projects suggests adaptability and a willingness to tackle diverse challenges. However, without specific psychometric or English test results, it's difficult to assess communication clarity, work attitude, stress handling, or team collaboration directly. The project descriptions are technically detailed, but lack explicit mentions of collaboration or stakeholder communication.