Mythri Shivakumar
Software Developer | Big Data Engineer | AI&ML Engineer
About
I’m Mythri — a curious CS grad student at UB with a knack for turning data into magic, and systems into symphonies. I spent two years at Infosys, where I honed my craft, and now I’m diving deeper into AI and innovation at UB. I’m passionate about building solutions that spark joy and solve real problems—whether it’s predicting air quality or crafting scalable backends. When I’m not coding, you’ll find me sipping coffee, exploring new tech trends, or dreaming up ways to make the world a smarter place.
I’m on the hunt for summer internship roles to fuel my growth, and I’m also looking ahead to full-time opportunities starting January 2026. Let’s create something amazing together!
Email: mythrish@buffalo.edu
GitHub: MythriShivakumar
LinkedIn: Mythri-Shivakumar
Experience
Software Development Engineer | Specialist Programmer @ Infosys
- Architected high-performance ETL pipelines using PySpark and AWS, processing data daily, reducing processing times by 90%.
- Migrated legacy Unix shell scripts to a cloud-native ETL platform, developing reusable components for validation, transformation, and aggregation.
- Engineered advanced preprocessing pipelines with filters, validators, and transformers, improving data quality and optimizing workflows.
- Led R&D initiatives on Python libraries and data engineering frameworks, driving platform scalability and maintainability.
- Collaborated with cross-functional teams in an Agile environment, aligning technical solutions with business goals, communicating complex concepts to stakeholders, and delivering iterative improvements.
Projects
Spam Classification with BERT and DistilBERT
Leveraged Hugging Face Transformers to implement BERT and DistilBERT models with linear probing for spam classification on the Enron Spam Dataset. Froze LLM weights and trained an MLP classifier head using PyTorch, achieving 98.36% accuracy.
Skills: Deep Learning, NLP, Hugging Face Transformers, PyTorch
Transformer Model for Text Classification
Implemented a Transformer model from scratch using PyTorch, incorporating multi-head self-attention, positional encoding, and feed-forward networks. Trained on a news dataset for multi-class classification (World, Sports, Business, Sci/Tech), achieving over 80% test accuracy through hyperparameter tuning and optimization techniques like ReduceLROnPlateau.
Skills: Deep Learning, NLP, PyTorch, Transformers
Autoencoders for Anomaly Detection
Designed a 1D Convolutional Autoencoder using PyTorch on the NYC Taxi Dataset, achieving 98% reconstruction accuracy for anomaly detection. Tuned hyperparameters and optimized architecture to identify urban mobility outliers, leveraging time-series analysis techniques.
Skills: Deep Learning, Autoencoders, Time Series Analysis, PyTorch
F1 Race Prediction Model
Developed a machine learning model to predict F1 race outcomes using historical race data, driver stats, and track conditions. Engineered features like lap time deltas and weather impacts, and tested models including Random Forest and Gradient Boosting. Achieved 85% accuracy in predicting top 5 finishers.
Skills: Machine Learning, Feature Engineering, Predictive Modeling
Air Quality Prediction Using LSTM
Built an LSTM model to predict air quality metrics using the AirQualityUCI dataset. Engineered features, handled missing data, and used GRUs to tackle vanishing gradients. Validated with rolling window and cross-validation.
Skills: Time Series Forecasting, RNN, LSTM, Deep Learning
Real Estate Sales Data Analysis
Analyzed 1M+ real estate transactions, engineered features like price per sq ft, and built Gradient Boosting, Random Forest, and XGBoost models. Deployed Tableau dashboards for trends and heatmaps.
Skills: Data Science, Machine Learning, Predictive Modeling, Data Visualization
VGG and ResNet for Image Classification
Implemented VGG and ResNet from scratch in PyTorch. Trained on benchmark datasets with custom weights, using batch normalization and dropout. Evaluated with accuracy, precision-recall, and Grad-CAM.
Skills: Deep Learning, CNN, PyTorch
Airbnb Price Prediction
Performed EDA on Airbnb data, cleaned and encoded features, and built Linear Regression, Decision Trees, and Random Forest models. Evaluated with R-squared and MSE for pricing insights.
Skills: Big Data, Data Analysis, EDA
Reinforcement Learning Vacuum Bot
Created a 6x6 grid-world for a vacuum bot using RL. Implemented SARSA with ε-greedy policy and n-step Double Q-Learning for stability. Fine-tuned hyperparameters for efficient navigation.
Skills: Reinforcement Learning, Machine Learning
EMNIST Character Recognition
Developed a CNN in PyTorch for EMNIST character classification. Used dropout, batch normalization, and data augmentation. Evaluated with confusion matrices and ROC curves.
Skills: CNN, Deep Learning, Optimization
Skills
Programming Languages
Software Development
ETL and Data Engineering
Machine Learning & AI
Frameworks & Libraries
Tools & Platforms
Methodologies
Resume
Want to dive deeper into my journey? Check out my full resume!
View ResumeContact
Let’s connect—I’d love to chat!
Email: mythrish@buffalo.edu
Phone: (716)-319-4689
LinkedIn: Mythri-Shivakumar
GitHub: MythriShivakumar