Daniel P. Findley

• Developed and productionalized Active Listening AI system to transcribe, summarize and extract facts using agentic approaches from meetings between clients and financial representatives.
• Led a transcription model comparison that reduced daily transcription costs by 60%, resulting in over one million dollars in annual savings.
• Led evaluation of small language models for extraction tasks, optimizing the balance of cost, efficiency, and accuracy.
• Designed and conducted large-scale experiments evaluating summarization outputs and LLM-driven data scraping pipelines.
• Built large-scale evaluation frameworks to assess LLM output quality across multiple use cases.
• Created agentic systems to monitor user feedback and surface answers to key questions from stakeholders.
• Developed content-based recommendation system leveraging Catboost and document embeddings to improve financial plans offered.
• Built a LangGraph agent embedding the recommendation system, providing models as tools to autonomously construct comprehensive financial plans.
• Developed automated plan generation model using a recurrent neural network to enhance financial advisor efficiency and plan quality.

• Created data flow diagrams for migrations of data from legacy systems to a unified data platform.
• Created, trained, tuned and evaluated machine learning models to categorize pages using both supervised and unsupervised methods.

• Worked with the public health division using census data in R and Datawrapper to visualize and explain the digital divide across the state of Wisconsin.

GPA: 4.00 / 4.00

GPA: 3.92 / 4.00 · Summa Cum Laude · Core Honors Program

• Built full end-to-end data pipeline with automated scraping from multiple football data sources, data manipulation, and quality control.
• Outputs goal, assist, clean sheet and match score predictions using XGBoost models and Monte Carlo match simulations.