Hi, I'm
Oliver Swack
Data Scientist
Applying research to real-world problems. Lifelong learner, dreamer, nature lover.
Experience
Data Scientist
I work at Forthright, a survey panel platform within Bovitz Inc. The team is small, which means the work is real. I own projects from ideation through deployment and contribute broadly across the company wherever applied data science and AI are involved. I serve as a source of expertise and direction on AI for the business. Beyond modeling, I also contribute on the data engineering side, building efficient data streams and pipelines that help the business better leverage its own data and sharpen existing methodologies. My work spans predictive modeling, agentic systems, and NLP, with a recurring thread of trying to better understand human behavior in the modern world. I'm used to wearing many hats, and I like it that way.
Undergraduate Research
My time as an undergraduate researcher was where I first learned what it means to take real ownership of something. Working alongside professors in the mathematics and earth sciences departments, I wasn't just following orders. I was responsible for figuring things out on my own, often in areas I had never touched before. Assisting in research pushed me to approach new problems with a critical lens, to sit with uncertainty, and to trust the scientific process. More than any specific method or tool, what I took away was a sense of intellectual maturity: how to ask better questions, how to investigate rigorously, and how to see a line of inquiry through. A paper based on research I contributed to during this time is slated for publication soon.
Projects
RAG Agent App
In ProgressLightweight retrieval-augmented generation layer for an agentic application. Uses Reciprocal Rank Fusion to blend multiple retrieval signals and surface higher-quality context for downstream LLM reasoning.
Agentic Survey System
Real-time multi-armed bandit system for dynamic survey optimization. Successfully completed pilot testing and deployed in production.
Particulate Matter CNN-LSTM
↗Deep learning model forecasting air quality at 80%+ accuracy. Established a 1TB+ async data pipeline on a computing cluster and accelerated training 400% with CUDA/GPU kernels.