Learn Data Skills
Beta
Paula Rescala

Paula Rescala

Certified

Job Hunting

N/A

Technologies

My Portfolio Highlights

My New Course

Understanding Cloud Computing

My New Course

Understanding Cloud Computing

Mindful analyst, decoding patterns to unravel the big picture.

My Work

Take a look at my latest work.

course

Understanding Cloud Computing

course

AWS Concepts

course

Introduction to Shell

course

Understanding Cloud Computing

My Certifications

These are the industry credentials that I’ve earned.

Data Scientist Associate

Data Scientist Associate

Data Literacy

Data Literacy

AI Fundamentals

AI Fundamentals

DataCamp Course Completion

Take a look at all the courses I’ve completed on DataCamp.

Courses

Introduction to SQL
Intermediate SQL
Understanding Cloud Computing
Introduction to Tableau

My Work Experience

Where I've interned and worked during my career.

École polytechnique fédérale de Lausanne | Aug 2023 - Jan 2024

Master's Thesis: An Observational Study on LLM Detection of Personalized, Convincing Arguments

Abstract: Since their inception, large language models (LLMs) have proven to (1) generate impressive, well-written content at the level of human ability and (2) perform well on a breadth of tasks. However, their ability to generate content so well and so rapidly may come at an ethical cost. It is feared that LLMs may, be it accidentally or at the hands of malicious actors, be used to craft personalized, convincing arguments that could easily include dangerous content like misinformation. Generating persuasive content is closely related to the task of identifying whether a particular argument is persuasive to a target user. This study analyzes state-of-the-art open and closed-source LLMs' ability to (1) detect convincing arguments, (2) determine people's stances on propositions given their demographics, and (3) detect whether arguments are persuasive enough to sway a person's opinion on a proposition based on their demographics. To do so, we prompt LLMs to perform classification tasks related to these three research questions zero-shot using content from a dataset collected from debates.org, a former online debate platform where users could vote and share which arguments and propositions they agreed with. We find that the best model can detect convincing arguments with ~60\% accuracy but, in general, most LLMs perform anywhere from slightly worse to slightly better than random (33%) at this and other detection tasks. Nevertheless, we also find that crowdsourced Amazon Mechanical Turk performance is worse than hypothesized and some LLMs are on par with it. Perhaps most surprisingly, we see that although lines of reasoning exist for why models predict the way they do, all models predict differently and succeed/fail at different instances of the same task. Overall, this work contributes an important new benchmark for testing LLM performance that so far follows the phenomenon of detection and prediction lagging behind generation.
Show More

Schindler Group | Feb 2023 - Aug 2023

Indoor Navigation System to Improve Elevator Accessibility for the Visually Impaired

This project aimed to improve accessibility of destination control elevator systems for blind and low vision people. In collaboration with another intern, we studied current interactions between visually impaired users and elevators and brainstormed and designed a new solution focused on the use of indoor beacon technology and haptic feedback. I personally was responsible for the entirety of the localization task and aided on the software development portion of the haptic feedback part. I achieved the following: • Collected a dataset of 7920 RSSI measurements from 3 BLE/LoRa beacons in an indoor environment by (1) coding drivers to receive measurements and send them to a Raspberry Pi and (2) developing a reproducible data collection protocol. • Used both my own and an existing dataset to train neural network models using contrastive learning and physics-informed frameworks in Azure ML which achieved results as performant as state-of-the-art. • Conducted a study evaluating the impact of hyperparameters, network architecture, dataset choice, and loss function on indoor localization results. • Designed and implemented a localization strategy of using an extended Kalman Filter to combine IMU position estimates using kinematics and beacon position estimates using the trained model. • Collaborated with 3 people on software development using Git and Azure DevOps by creating and reviewing pull requests amounting to 4000+ lines of merged code including over 30 unit tests. • Led some and participated in all sprint planning sessions (including task planning, time estimation, and retrospective meetings) as part of SCRUM project management. • Supervised a new student after project completion to test my final ideas for further improvement.

École polytechnique fédérale de Lausanne | Sep 2022 - Jan 2023

Semester Project: A Human-Centered Approach to Understanding Local News Consumption

Abstract: Studies conducted in countries and cities across the world have shown that patterns in news consumption differ by demographic factors such as age, race, socioeconomic status, and more. This is the case for news consumption at every level: the local, regional, national, and international. Nevertheless, not much research goes beyond simply presenting such differences in consumption habits. There exists a lack of understanding as to why certain disparities exist between demographic groups. Moreover, news is often seen as a ground truth possessing inherent value rather than as a product that should meet the demands of its readership. Thus, this project aims to fill two gaps in news consumption research: (1) Studying the migrant community's news consumption habits along with identifying their sentiments, perspectives, and needs when it comes to interacting with the media and (2) Analyzing published news content systematically using natural language processing techniques to unearth whether the identified community needs are being met. We invited five French-speaking members of the migrant community living in Lausanne, Switzerland to participate in a focus group aimed at extracting insights about the media needs this community has and the sentiments they feel towards news. Then, guided by the discussion, we performed topic modeling, information retrieval, sentiment analysis, and text readability analysis of 2666 web-scraped articles collected from Lausanne Cités, a hyper-local news platform, and connected our analysis to the focus group's qualitative results. Overall, our work attempts to bridge the gap between news platforms' content and readers' needs.

Sandia National Laboratories | Aug 2020 - Mar 2021

Research And Development Intern

Continuing work from summer O-REU at TAMU, sponsored by Sandia National Labs. Using reinforcement learning/machine learning approach for developing suitable discretizations for use in mesh-free numerical methods. Additionally using optimization techniques such as nonlinear integer programs for solving a clustering problem. Collaborating remotely with my mentor.

Rice University D2K Lab | Aug 2020 - Jan 2021

Senior Design Project: Post-Operative Arrhythmia Detection in Pediatric Cardiac Patients

Partnered with Texas Children's Hospital and worked with 2 other undergraduates and 2 PhD students to create an arrhythmia detector that minimized false alarms, which currently result in severe physician fatigue. Cleaned, explored, and feature engineered data from 4 cardiac signals in patients with congenital heart disease using signal processing techniques which then fed into a deep network.

Tapia Center for Excellence and Equity in Education | Nov 2020 - Dec 2020

Tapia Camp Instructor

I served as an instructor for a data visualization camp for underserved eighth and ninth graders. I taught math-focused lessons and helped students create a data visualization presentation on a societal problem of their choosing. I also served as a mentor to answer their questions about college and encourage them to pursue their interest in STEM.

Texas A&M University O-REU | May 2020 - Aug 2020

Computational Solid Mechanics Researcher

N/A

Rice University | Apr 2018 - May 2020

Rice Bhangra Captain

Bhangra is a traditional folk dance from Punjab, India. As a captain, I choreographed competitive and creative routines that incorporated aspects of both the traditional folk dance and modern movement and music. I led the team in practices for 6 hours a week along with my co-captain and sought performance opportunities at cultural events, competitions, weddings, etc. I also collaborated with the publicity chair, socials planner, competition coordinator, and treasurer, using past successful experiences in these roles as a guide.

IMTEK, University of Freiburg | May 2019 - Aug 2019

Research Intern

I developed and refined post-processing technology for 3D-printed fused silica glass parts. I innovated the experimentation process by 3D-printing resilient structures that aided in dip-coating glass parts. I also worked on a task force to produce samples that were exhibited at Rapid.Tech.Fab.con 3.D. The samples contributed to the team winning first place at the conference for "Start-up of the Year."

Rice University | Oct 2018 - May 2019

Undergraduate Research Assistant in Mechanical Engineering

N/A

University of South Florida | May 2018 - Sep 2018

Undergraduate Research Assistant

I proposed possible nonlinear metamaterial designs that would maximize the second harmonic generation efficiency of the material and collected data on this efficiency through simulations on CST Microwave studio.

My Education

Take a look at my formal education

Master of Science in Data ScienceÉcole polytechnique fédérale de Lausanne | 2024
Bachelor of Arts in Computational and Applied MathRice University | 2021

About Me

Paula Rescala

I am a Data Scientist passionate about working on large-scale, impactful projects in cross-functional teams. I’m eager to transition into the industry, applying my expertise to real-world challenges and making a positive difference.

Powered by

  • Work
  • Certifications
  • Courses
  • Experience
  • Education
  • About Me
  • Create Your Data Portfolio for Free