Eric (Yunjae) Soderquist

A free, AI-driven platform providing real-time voice translation across 50+ languages.

• Led all coding and machine learning development, creating a custom multimodal language model utilizing Reinforcement Learning from Human Feedback (RLHF) and divergence training, resulting in high-quality, context-aware translations.
• Architected and maintained cloud-native infrastructure using TypeScript, Next.js, Supabase, and Azure; deployed models on Azure OpenAI service and secured cloud funding through a Microsoft partnership.
• Implemented low-latency, real-time interactions by leveraging Azure's infrastructure to ensure high availability and scalability for seamless user experiences.
• Managed end-to-end engineering, data science, and data engineering efforts, from model training and fine-tuning to infrastructure optimization, enabling translations adaptive to sociocultural and emotional contexts.
• Achieved over 1,000+ active users, providing accessible, real-time translation support to immigrants, ESL learners, travelers, and language learners.
• Pioneered innovations in AI translation using advanced machine learning techniques, creating a scalable, cloud-native solution that democratizes access to high-quality language tools.

Self-hosted media streaming infrastructure with advanced features:

• Content-based recommender systems using unsupervised clustering and NLP
• Metadata-rich UI for enhanced discoverability
• Automated content ingestion pipelines
• User-driven content acquisition portal
• Adaptive bitrate streaming with client-side network analysis
• High-throughput GPU-accelerated transcoding
• Premium audio-visual codec integration (Dolby Atmos, Vision, HDR10+)
• IPTV integration
• Automated polyglot subtitle and audio track retrieval

25+ Monthly Active Users, exceeding commercial streaming quality benchmarks

Development of unsupervised neural network models focusing on autoencoder architectures employing backpropagation algorithms for efficient data encoding and reconstruction tasks.

• Implemented both single-layer and two-layer autoencoders in Python, optimizing algorithmic efficiency and scalability within high-dimensional data contexts
• Conducted in-depth exploration of non-linear dimensionality reduction techniques and feature extraction methodologies, utilizing sigmoid activation functions and gradient descent optimization
• Engaged in rigorous hyperparameter tuning to enhance model convergence and minimize reconstruction loss, adhering to computational standards in unsupervised learning paradigms
• Leveraged Python's scientific computing libraries including NumPy for high-performance numerical operations, Scikit-learn for machine learning utilities, and Matplotlib for data visualization
• Contributed to advanced neural network modeling within the context of cognitive psychology and neuroscience, aligning with the academic objectives of PSYC 489: Neural Network Modeling Lab
• Underscored the applications of autoencoders in dimensionality reduction, anomaly detection, and feature learning, enriching the discourse in unsupervised machine learning and data representation

Construction and analysis of an advanced Hopfield Network framework to simulate associative memory and pattern retrieval in neural systems, employing recurrent neural network architectures with symmetrically weighted connections and energy minimization principles to achieve stable memory states.

• Focused on accurate simulation of network dynamics based on the original formulations by John Hopfield, exploring the network's capacity for content-addressable memory and its applications in optimization problems
• Implemented a class-based structure in Python for modularity and extensibility, incorporating methods for network training on multiple memory patterns, state updates with synchronous and asynchronous options, and precise energy calculations to analyze convergence behaviors
• Leveraged NumPy for efficient numerical computations and Matplotlib for visualizing energy landscapes and network states
• Enhanced understanding of cognitive functions and contributed to computational neuroscience studies within the context of PSYC 489: Neural Network Modeling Lab
• Offered insights into memory models, pattern recognition, and the underlying mechanics of neural associative processes

Sentiment classification on large-scale movie review corpus using optimized recurrent neural architectures.

• Employed sequential memory modeling for temporal dependency capture in natural language
• Performed hyperparameter tuning via grid and stochastic search methodologies for performance maximization
• Leveraged distributed representations and transfer learning from pre-trained word embeddings
• Implemented within TensorFlow/Keras ecosystem with auxiliary data manipulation via Pandas and Scikit-learn
• Contributed to affective computing research in human-computer interaction and opinion mining domains

Stochastic simulation framework for multi-agent decision-making in constrained combinatorial environments.

• Leveraged Monte Carlo methods for strategy optimization in dice-based games
• Conducted statistical inference on high-dimensional gameplay metrics via hypothesis testing and multivariate regression
• Performed data-driven insights extraction through exploratory data analysis and advanced visualization techniques
• Implemented in Python with NumPy/Pandas for numerical computing and Matplotlib/Seaborn for graphical representation
• Contributed to computational game theory and strategic decision analysis in uncertain, rule-bound domains