Welcome to the Portfolio ofAkshat Karwa
BS & MS Computer Science @ Georgia TechData Science & Machine Learning Enthusiast
Lets Connect!
About Me
I am a computer scientist passionate about data science and innovative problem-solving, currently pursuing my MS in Computer Science at Georgia Tech. With expertise in algorithmic optimization, machine learning, and statistical modeling, I develop systems that transform complex data patterns into actionable intelligence. My work combines rigorous analytical methods with creative problem-solving to build scalable, user-centric applications that bridge the gap between technological innovation and real-world utility.
I maintain an insatiable curiosity and embrace continuous learning, adapting my skillset to tackle increasingly complex computational challenges. When I'm not coding, you'll find me exploring new technologies or collaborating with developers on open-source projects. Driven by the belief that thoughtfully designed systems can create cascading positive impacts, I am committed to democratizing technology access and building solutions that make a meaningful difference in people's lives.
Let's build something impactful together.
Selected Projects
BuzzAI - Course Discovery System - Georgia Tech
Software Engineer
- Built course recommender using transformer models (MiniLM, DistilRoBERTa) and FAISS, with custom filtering, collaborative profiling, and Llama model for generating topic descriptions, achieving 84.1% accuracy in course matching.
- Engineered data pipeline with custom web scrapers (BeautifulSoup, pdfplumber) and vector embedding architecture, processing 2,000+ courses and enabling adaptive recommendations with 90%+ relevance through user preference learning.
COVID Forecasting with Exogenous Data Integration
Data Scientist
- Engineered an epidemic spread forecasting system utilizing state-of-the-art models (SARIMA, Prophet, LSTM, RNN, SI, TBATS), integrating exogenous data across 1,674 days, reducing prediction error by 24% over traditional approaches.
- Optimized performance through strategic feature integration and combination analysis of 38 features including vaccination rates, mobility, health and economic indicators, achieving 31% improvement in prediction accuracy.
Spotify Song Genre Classifier & Popularity Predictor
Machine Learning Engineer
- Analyzed Spotify data encompassing 38,000+ songs to develop a predictive model, achieving 87.3% accuracy in genre classification using Random Forest (F1 score: 0.88) and 96% accuracy in hit prediction.
- Conducted data pre-processing, explored dimensionality reduction techniques like PCA, t-SNE, and UMAP, applied ML algorithms, and created visualizations to uncover patterns in music genre and popularity trends.
Voice Cloning Using Deep Learning
Machine Learning Engineer
- Built a voice cloning system using YourTTS and TortoiseTTS, achieving exceptional voice replication (Mean Cosine Similarity: 0.982 & Mean MSE: 0.019) through architectural optimizations and effective fine-tuning.
- Refined a speaker encoder through multi-stage training on LJSpeech and VCTK datasets, achieving 0.4036 Mel-spectrogram similarity on 11 unseen speakers, reducing spectral distortion and enhancing synthesis naturalness.
Automated Algorithm Design
Software Engineer
- Developed the EMADE framework, implementing a multi-island model with diverse objectives and dynamic migration topology that improved population diversity by 38% phenotypically and 11% genotypically, outperforming algorithms like NSGA-II and MOEA/D.
- Optimized framework performance, achieving a 18% improvement in convergence speed, 22% better objective coverage, and reducing computational overhead by 27% on the Georgia Tech PACE cluster.
BeeHired - Job Aggregation Portal
Software Engineer
- Created high-performance web crawler processing listings at 1.04s/post with 0.14s parsing speed for 100+ posts, and optimized storage to 0.08s/entry, through Selenium WebDriver implementation.
- Architected full-stack job platform using React/Express/MySQL processing 10,000+ posts in 24 mins, featuring 0.07s query response time, ghost job detection, and user-friendly multi-parameter filtering interface for personalized job search.
Real-Time Algorithmic Trading Platform
Software Developer
- Developed an algorithmic trading platform integrated with Kotak Mahindra Bank API, enabling automated trade execution with average latency under 50ms, and reducing manual intervention by 90%. Implemented robust error handling to ensure continuous operation during market hours.
- Conducted comprehensive strategy back-testing using 5 years of historical market data to optimize performance and adapt to current market conditions, and optimized algorithm performance through efficient data handling techniques.
Airline Database Management System
Database Engineer
- Implemented MySQL procedures, views, and structures to administer passenger/pilot profiles and movements, flight assignments, takeoffs/landings, delays, ticket purchases, route creation and boarding/disembarking.
- Analyzed scenarios, handled dataset irregularities, addressed edge cases and tested via simulation cycles.
Road Crossing Mobile Game
Mobile Developer
- Developed a feature-rich mobile game implementing gameplay mechanics, player functionalities, and game assets.
- Integrated multiple screens and optimized UI/UX for seamless performance and an engaging experience across devices.
Coursework Constellation
Data Visualization Engineer
- Built an interactive D3.js visualization showing 35+ courses that I have taken at Georgia Tech, where nodes represent courses and edges show course relationships, with color gradients reflecting course depth/prerequisites.
- Created a dual-panel layout: left panel displaying standalone general education courses with no prerequisites, while the right panel mapping the interconnected technical curriculum through directed edges showing course dependencies.
Machine Learning Specialization
ML Developer
- Designed and implemented diverse ML algorithms including neural networks, SVMs, and recommender systems using GNU Octave, achieving optimal model performance through systematic hyperparameter tuning.
- Built and optimized supervised and unsupervised learning models while mastering fundamental ML concepts through hands-on implementation of clustering, gradient descent, and backpropagation algorithms.
JavaScript Applications
Full Stack Developer
- Engineered full-stack web applications including a real-time food ordering system, an interactive budget management tool, and a live stock trading platform using React, Node.js, and MongoDB.
- Implemented modern development workflows with NPM, Parcel bundling, and Babel for optimized performance, reducing load times by 40% and enabling seamless ES6+ compatibility.