Projects

Data visualization project analyzing generational and gender differences in social and political attitudes using ESS survey data.

Developed iLLuMinaTE, a LLM pipeline that delivers actionable, theory-driven feedback to students, outperforming traditional explainers with 89.52% student preference across diverse online courses.

Chain-of-Thought training of an LLM for university-level MCQA using a pipeline with SFT, DPO, and RAG

Examined over 20,000 movie plots using NLP and Network Analysis to explore relationships and interactions between main/side characters and between genders.

Simulated the launch of a MVP product, starting with product discovery via interviews and market research, followed by the definition of product vision and roadmap, leading to app development and final product launch.

Developed a RLAIF pipeline by replicating Anthropic's 2021 publication 'Constitutional AI' aimed at ethically guiding the Llama-7b-chat model through a set of predefined principles.

Topic modeling and NER on a historical dataset containing 600+ interview transcripts of individuals who experienced the Manhattan Project. Through our methods, we discovered the individuals and topics most remembered and how they are connected.

Designed and developed a functional website allowing users to generate AI-driven variations of Montreux Jazz Festival posters, with the goal of promoting artist discovery and appreciation.

With my team, I created ML models for music generation, including rhythm, melody, and musical form. Despite being new to this domain, I invested significant effort into metric design and model selection. The project also led me to explore techniques like GANs, Markov chains, and GMMs.

Data mining exploration of a Spotify podcasts dataset through text network analysis that resulted in new inspiring hypotheses for further research.

Following a design process from research to prototyping, we designed Venn, a music streaming app. Venn lets users discover their preferences with highlighted music tags. This new musical awareness encourages a personalized discovery, as users dynamically engage with the improved recommendation algorithm.

Clothesline is a unique community-driven app that shifts the focus from buying and selling second-hand clothing to sharing. We designed this application using a hands-on, workshop-based approach, employing techniques such as user journeys, state of the art research, personas, and more.

Designed a new citizen-driven platform, following the double diamond model. We brainstormed, researched, and selected our core idea, validating it through user interviews and a social media survey. Finally we designed a digital prototype in Figma and performed user testing.