Nicholas Konovalenko

Technical Expertise

Machine Learning at Scale

Architected and launched deep learning recommendation models across critical Roblox discovery surfaces. Delivered multiple production releases that moved the needle on key engagement and revenue metrics.

MSc Artificial Intelligence (Boston University) • BSE Computer Science (University of Michigan)

100M+ Users Impacted Stat-Sig Metric Improvements Multiple Production Launches

Python

PyTorch

Spark

Scala

AWS

Kubernetes

SQL

Airflow

Featured Projects

🔇 Silencing the Foul Utterances

Real-time audio censoring pipeline that detects profanity in streaming audio and inserts censor beeps on-the-fly. Built a training set by multithreading Google Text-to-Speech over a labeled Kaggle Dota chat corpus, converting audio to spectrograms in NumPy/Librosa and training PyTorch CNN/RNN models for word-boundary profanity detection. Introduced the idea of "Tall & Skinny" convolutions of varying widths to cature the positional information.

class ProfanityDetector(nn.Module): def __init__(self): super().__init__() self.conv = nn.Sequential( nn.Conv2d(1, 16, 3, padding=1), nn.ReLU(), nn.MaxPool2d(2), nn.Conv2d(16, 32, 3, padding=1), nn.ReLU(), nn.AdaptiveAvgPool2d(1), ) self.head = nn.Linear(32, 2) def forward(self, mel): # mel: [B, 1, n_mels, time] x = self.conv(mel) x = x.flatten(1) return self.head(x)

Audio ML Regression Model 212k+ Generated Audio Samples Multithreaded Google TTS Pipeline

PyTorch Signal Processing Librosa Google TTS

View Source Code

🎯 Deep Learning Go AI

Competition-grade Go playing bot combining Monte Carlo Tree Search with neural networks. Implemented naive, minimax, and MCTS algorithms to compare performance. Used Zobrist hashing to compact state keys and accelerate MCTS, and applied Alpha-Beta pruning to further reduce the search space. Achieved amatuer Dan level performance in online tournaments, demonstrating significant improvement in game theory understanding over naive algorithms.

class AlphaGoZero: def mcts_search(self, state, simulations=800): # Advanced MCTS with neural network guidance for _ in range(simulations): leaf = self.select_leaf(state) value, priors = self.network.predict(leaf) self.backup(leaf, value) return self.get_action_probs(state)

Amateur Dan Level Performance Monte Carlo Tree Search AWS Deployment

PyTorch Game Theory Zobrist Hashing Alpha-Beta Pruning

View Source Code

Professional Experience

2023 - Present

Machine Learning Engineer

Roblox Corporation • San Mateo, CA

MLE on the Discovery Home Modeling Team for the recommendation system serving 100M+ daily active users across Home, Search, and Charts. Spearheading ML initiatives that directly impact platform revenue, user engagement, and retention metrics. Leading cross-functional collaboration efforts with Product, Engineering, and Data Science teams.

Designed and deployed multiple production ML models with statistically significant metric improvements
Engineered 20+ new features that improved user engagement and content diversity metrics
Improved ML understandability with 10+ analysis tools to optimize oncall debugging efficiency
Applied Computer Vision techniques to evaluate content quality

PyTorch Spark Scala Kubernetes Airflow A/B Testing ML Explainability

Summer 2022

Software Engineer Intern

Microsoft Corporation • Bellevue, WA

Developed enterprise-grade automation tools for internal teams, focusing on secure data processing and workflow optimization. Built full-stack solutions handling sensitive user data across distributed systems.

Architected React-Redux frontend with C# backend processing 100K+ daily transactions
Implemented secure data pipelines with Kusto for real-time analytics
Reduced manual processing time by 60% through intelligent automation

TypeScript React C# Azure Kusto

Summer 2021

Software Development Engineer Intern

Amazon Lab126 • Sunnyvale, CA

Built high-performance automated testing platform for Computer Vision and Camera teams. Optimized data processing pipelines for edge computing environments with strict resource constraints.

Engineered automated testing platform processing 0.5 GB/s on ARM architecture
Optimized multithreaded data pipeline reducing CPU usage by 40%
Implemented edge-optimized algorithms for real-time computer vision processing

Java Computer Vision ARM Multithreading

Let's Build Something Amazing

Interested in discussing Machine Learning, recommendation systems, or potential collaborations? I'm always excited to connect with fellow enthusiasts and explore innovative projects.