ML Portfolio
From my beginnings as a Quant UX researcher, I've charted a transformative journey toward becoming a Research Scientist deeply rooted in machine learning. My background in user experience and quantitative research has infused me with a distinctive vantage point, allowing me to bridge the chasm between intricate technical models and their real-world applications. With users as the bedrock of my methodologies, I craft machine learning models that are technically robust and intuitively aligned with user-centric solutions. My diverse portfolio stands as a testament to my dedication, expertise, and the unique blend of research and technology that I bring to the table.
Research Librarian
The Research Librarian project is an AI-powered index designed to improve internal accessibility and search capabilities for large amounts of UX and CI research data. Its groundbreaking algorithm transforms how research data is organized, categorized, and accessed. The system autonomously selects the most suitable query engine to provide users with precise and insightful data points.
To ensure optimized performance and a seamless user experience, the project leverages a robust multi-container and multi-build architecture. Each Docker container is tailored to a unique query engine, increasing data retrieval speed and deepening the richness of results. This innovative approach establishes a harmonious blend of advanced technology and meticulous research organization.
Modular Survey Analysis System
I developed the Modular Survey Analysis System, an evolution of the initial survey report generator. This new system streamlined the process by moving away from a static, hard-coded approach to a more dynamic, modular method, ensuring efficient categorization of diverse questions. A significant part of this project was integrating context-aware logic and enhancing the data interpretation process. One of the key components I'm particularly proud of is the autonomous clustering algorithm designed for open-ended responses. Currently, in the patenting process, this algorithm demonstrates a thoughtful approach to addressing a common challenge in survey analysis.
The development process presented several challenges. One such challenge was the inconsistency in the survey data due to its diverse nature. I addressed this by incorporating precise preprocessing techniques, ensuring data uniformity. Additionally, I encountered surveys with complex logic and dependencies. To tackle this, I designed a logic parser that effectively managed these intricacies. The open-ended responses, inherently subjective, were another challenge. My solution, the clustering algorithm, effectively categorizes and describes these responses, underscoring my dedication to creating efficient and reliable solutions.
Customer Support Bot
I aimed to improve customer support with the help of cutting-edge AI technology. I developed an autonomous agent that could provide prompt and concise responses to user queries, thereby reducing the workload on our call center and improving self-service for our users.
I used advanced AI technologies like embeddings and GPT-3.5 to power the agent. I also created a comprehensive index of all the articles on a CS site, in this case, Roku, to ensure users could easily access the most relevant information. The development process involved indexing all the support site articles and creating a library catalog. Next, I used embeddings to search for the most relevant content and then utilized the advanced AI model of GPT-3.5 to generate a response. This has greatly improved the efficiency of our customer support system, allowing us to provide faster and more accurate solutions to our users.