I'm a data scientist and ML engineer starting an MS in Data Science at NYU. Most of my work so far has been production machine learning: retrieval and reranking systems, vector search, and data infrastructure at scale. What I want to work on next is narrower. I care about how AI systems reason, and I think the most useful ideas about that come from cognitive science.

Portfolio, resume, and writing: lcpizani.github.io

My main question is whether we can build models that reason more like people do: models that generalize from a few examples, compose what they know into new situations, and hold up when the problem shifts. Cognitive science has spent decades describing how human reasoning works and where it breaks. I want to use that as a guide for training and evaluating models, rather than treating better benchmark scores as the goal.

Related things I keep coming back to:

• Trustworthy AI. Grounding model outputs in sources you can check, measuring calibration honestly, and knowing when a deterministic computation should override a generative one.
• Evaluation. Telling the difference between a model that has a capability and one that has learned the shape of the test.
• Structured representation. Knowledge graphs, entity linking, and other ways of giving a system something explicit to reason over.

I'm early in this shift. The projects below are where my current work already touches these questions.

• Languages: Python, R, SQL, Java, JavaScript, TypeScript
• ML and AI: PyTorch, scikit-learn, Sentence Transformers (SBERT), NetworkX, LangChain, RAG pipelines, prompt and agentic workflow design
• Data and infrastructure: PostgreSQL, ChromaDB, FastAPI, Docker, GCP, AWS, BigQuery, pandas, NumPy
• Analytics: Tableau, Looker Studio, large-scale hierarchical clustering, Monte Carlo simulation

BS in Data Science and Analytics and Mathematics, St. Ambrose University (4.0 GPA, Summa Cum Laude, 2026), with minors in Computer Science and Statistics. Member of Pi Mu Epsilon. Former NAIA scholar-athlete in men's soccer.

• Website: lcpizani.github.io
• LinkedIn: Lucas Pizani