AI Specialist - Data Scientist, ML

AUGUSTIN KIM

Los Angeles, CA 213-***-**** ************@*****.*** linkedin.com/in/augustin-kim github.com/ajaykim94 PROFESSIONAL SUMMARY

Data Scientist and AI/ML practitioner with 7+ years of high-stakes analytical experience and a UC Berkeley M.S. in Information and Data Science. Skilled in computer vision, NLP, deep learning, geospatial analytics, and model evaluation using Python, PyTorch, TensorFlow, Scikit-learn, SQL, and cloud-based ML workflows. Built applied projects across emotion detection, kelp canopy segmentation, churn prediction, and wildfire risk analytics, translating complex model outputs into clear recommendations for technical and non-technical stakeholders. Inactive TS/SCI with Poly. SKILLS

• Languages/Databases: Python, SQL, NoSQL, Pandas, NumPy, Scikit-learn, PostgreSQL, Vector Databases.

• AI/ML & Deep Learning: PyTorch, TensorFlow, Hugging Face, Computer Vision, NLP, CNNs, Transformers, RAG, LLMs.

• Data Analysis & Viz: Tableau, Matplotlib, Plotly, Seaborn, ETL, Geospatial Analysis, Remote Sensing, Research Methods.

• Professional: Model Evaluation, Feature Engineering, Error Analysis, Responsible AI, Documentation, Stakeholder Communication.

EDUCATION

UNIVERSITY OF CALIFORNIA, BERKELEY

Master of Information and Data Science (M.S.) Dec 2025 Bachelor of Arts (B.A.) in Data Science May 2023 TECHNICAL PROJECTS

Emotion Detection System (Natural Language Processing)

• Built a text-based emotion recognition system for online communication, classifying tweet/sentence input into joy, sadness, anger, fear, love, and surprise using NLP and deep learning.

• Trained and compared Naive Bayes, Deep Averaging Network (DAN), CNN, BERT, and ensemble models using TF-IDF, tokenization, embeddings, class weights, confusion matrices, accuracy, and macro-F1 evaluation.

• Improved performance from an 82% Naive Bayes baseline to 91% CNN accuracy and 95% ensemble accuracy while analyzing class imbalance, label noise, missing context, sarcasm, and emotion-pair misclassifications. ForSEAble Kelp Canopy Detection (Remote Sensing & Computer Vision)

• Developed an ML workflow using Sentinel satellite imagery and environmental data to detect and estimate kelp canopy coverage across California coastal regions.

• Trained and evaluated segmentation/deep learning approaches including DeepLabV3+ and SegFormer, addressing class imbalance, thresholding, and spatial validation challenges.

• Created reproducible data pipelines for geospatial preprocessing, patch metadata, train/validation/test splits, and model comparison to support kelp restoration decision-making. Customer Churn Prediction & NLP Strategy (ML & Text Mining)

• Built an end-to-end churn prediction pipeline using Python, feature engineering, and Gradient Boosting/XGBoost models, achieving 0.83 ROC-AUC.

• Applied NLP/text mining to customer feedback and retention signals to identify churn drivers and segment high-risk customers for targeted interventions.

• Used SHAP analysis and threshold tuning to explain model behavior, quantify business trade-offs, and communicate recommendations to stakeholders.

PROFESSIONAL EXPERIENCE

UNITED STATES ARMY Various Locations

Signals Analyst Jul 2012 – Nov 2019

• Analyzed and interpreted complex intelligence data to identify critical patterns, mirroring high-stakes data science workflows under strict accuracy requirements.

• Translated technical findings into concise briefings for non-technical leaders, enabling rapid decisions in time-sensitive operational environments.

• Managed large-scale data projects independently while maintaining strict security protocols, documentation standards, and data accuracy controls.

• Modernized reporting templates to structure raw data, improve trend visibility, and strengthen organizational ability to identify emerging threats leading to the capture of over 500 insurgents.

Contact this candidate