Computer Science Student @ Oregon State University | GPA: 3.97/4.0 | Sophomore (2nd Year)
I'm passionate about Machine Learning, Computer Vision, and Backend Engineering. Currently working on ML research at Warren Lab, focusing on image classification and object detection for biological research applications.
class KonstantinSavkin:
def __init__(self):
self.location = "Corvallis, OR"
self.education = "Oregon State University - Computer Science"
self.gpa = 3.97
self.current_focus = ["Machine Learning", "Computer Vision", "Backend Engineering"]
self.current_role = "Laboratory Assistant @ Warren Lab"
def get_interests(self):
return {
"AI/ML": ["Computer Vision", "Image Classification", "Object Detection"],
"Backend": ["RESTful APIs", "Database Design", "System Architecture"],
"Tools": ["TensorFlow", "PyTorch", "Flask", "SQLAlchemy", "SQL"]
}Team Project: PolyDebate | DevPost
🏆 Awarded 3rd Place Overall out of 112 participants. Prize: Resume circulation to the Polymarket engineering team.
AI-powered debate platform integrating 100+ AI models with real-time prediction market data and TTS narration.
My Contributions (Backend/API Focus):
- Polymarket Integration (Critical):
- Engineered the ingestion pipeline for Polymarket's CLOB (Central Limit Order Book) API to fetch real-time event probabilities.
- Implemented logic to inject live market sentiment (e.g., "65% Yes") into LLM contexts, forcing AI debaters to ground arguments in current market reality.
- Backend Architecture: Built the core Flask orchestrator managing the flow between 3 external APIs (Polymarket, OpenRouter, ElevenLabs) and the frontend.
- Real-time Streaming: Implemented Server-Sent Events (SSE) to stream debate chunks and audio URLs instantly to the client, reducing perceived latency to near-zero.
- Database Design: Designed the schema to store debate history, user votes, and market snapshots for historical analysis.
- Delivered: 24-hour development sprint from concept to working product.
Position: Laboratory Assistant | Computer Science Department | February 2025 - Present
Trajectory Tracking & Behavioral Analysis System:
- Developed real-time trajectory tracking system for behavioral analysis of biological specimens during movement
- Implemented computer vision pipeline to classify specimens and visualize movement paths with color-coded trajectories
- Created automated visualization tool that tracks spatial coordinates and renders continuous movement patterns
- Enhanced research workflow by enabling simultaneous classification and spatial-temporal behavioral analysis
ResNet-50 Classification Model:
- Engineered ResNet-50 deep learning model achieving 98.24% test accuracy on 3-class biological specimen classification
- Developed custom training pipeline with data augmentation (rotation, flipping, color jittering, contrast adjustment) on 850+ image dataset
- Applied advanced tiling techniques for fine-grained feature detection in microscopic image analysis
- Improved model accuracy from initial baseline to 93% through systematic hyperparameter tuning and architecture optimization
- Curated balanced dataset (~450 samples per class) ensuring robust model generalization across morphological variations
- Authored technical report documenting ResNet architecture, training methodology, and evaluation metrics
Detection Model Project:
- Annotated 1,000+ images across 3 classes for object detection training
- Automated labeling pipeline achieving 99.6% time reduction (20s → 74ms per image)
- Implemented YOLO-based detection model for automated biological specimen identification
BeeMachine Parser:
- Designed and implemented parser for large-scale ML dataset processing
- Automated error analysis and misclassification flagging for model improvement
- Enhanced research workflow efficiency through data pipeline automation
🤖 PolyDebate - AI Debate Platform
QuackHacks 2025 Winner (3rd/112) | Team Project | Backend + AI
AI-powered debate simulation platform integrating 100+ AI models with real-time text-to-speech narration
Technologies: Python, Flask, SQLAlchemy, TypeScript, Next.js, OpenRouter API, ElevenLabs API
My Backend Contributions:
- Flask backend architecture with 8-table database schema
- RESTful API design (15+ endpoints) with JWT authentication
- External API integration (OpenRouter, ElevenLabs, Polymarket)
- Real-time SSE streaming implementation
- Database design and ORM modeling with SQLAlchemy
Stats: 111 commits | 3 contributors | 24-hour development
Complete SQL Database Design | Solo Project
Comprehensive 8-table relational database for university information with advanced SQL queries
Technologies: SQL, SQLite, dbdiagram, DBeaver
Features:
- 8-table normalized database schema
- 12 example queries demonstrating SQL mastery
- Advanced techniques: CTEs, window functions, JOINs, transactions
- Complete documentation with visual database diagram
Status: ✅ Fully documented and complete
Educational Project | Yandex Lyceum (2021)
Blindfold chess training app with Telegram bot and Yandex Station voice interface
Technologies: Python, Telegram Bot API, Yandex Voice API, Chess Engine
Features:
- Dual-platform support (Telegram + Voice)
- Chess engine with move validation
- Hint system with board visualization
- Voice-controlled hands-free gameplay
- 🧪 Warren Lab ML Research: Image classification and object detection for biological research
- 🎓 University Coursework: Data Structures, Algorithms, Database Systems
- 💡 Side Projects: Building ML model deployment pipelines and backend systems
- Advanced Computer Vision and Deep Learning architectures
- Distributed systems and scalable backend design
- Cloud deployment for ML models (AWS SageMaker, Azure ML)
- Advanced SQL optimization and database design patterns
- Email: [email protected]
- LinkedIn: linkedin.com/in/ksavkin
- GitHub: @ksavkin
- Location: Corvallis, OR
I'm actively seeking Summer 2026 internships in:
- Machine Learning / AI Engineering
- Computer Vision / Research
- Backend Engineering (Python/Flask/FastAPI)
- Data Engineering / MLOps
⭐️ From ksavkin | GPA: 3.97/4.0 | Oregon State University CS