About Me

Hi, I’m Michelle. I’m a design-oriented data scientist.

I’m obsessed with how people interface with technology and how to design products that serve us. I love working at the intersection of datascience product and UXUI to create meaningful online and offline experiences.

I obtained my MSc in Electrical engineering from The University of Southern California specializing in statistical modeling and optimization methods, machine learning for data analytics, and engineering project management. I completed my BSc in Computer Science from The University of Texas at Austin.

I’ve interned as a technical product manager at: quantum and quantum

While at ,I served as a User Experience Design Teaching Assistant for DSCI556

While at , I founded the UT Quantum Collective

I also mentored as a Data Science Instructor for The Coding School

Catch me at the 🧗🏻‍♂️gym coaching, out at a ☕️ shop, or making 🎨

Currently Fun

TechnoTaggr

Exploring audio signal processing with Essentia Tensorflow models to develop a music analysis tool which can analyze a folder containing music of various formats and add features metadata to audio files useful for categorizing and grouping tracks for DJing.

Github | Ploty Dash

Link to original

CRAFT - Mediation Chatbot

Developing a mediation chatbot for chat interface demo using the CRAFT model as a detection agent derailement, and a LLM augment with mediator role as a response agent using Thomas-Kilmann conflict resolution styles. Scenario is for dispute resolution between buyer and seller.

Github

Link to original

A brief History of me

Work History

--- 
config:
 theme: 'base' 
 themeVariables: 
   primaryColor: '#FAE072'
   primaryTextColor: '#36454F'
   secondaryColor: '#E0B3B3'
   tertiaryColor: '#D8F3DC'
   
---
timeline
    title Michelle's Timeline and OKRs
    section UT Austin | 2018 - 2023
        UT Quantum Collective 
        : Qiskit Fall Fest
		: S.Q.U.I.D. 2023
		: Quantum Coalition Partnership
        Microsoft
        : Azure Quantum Partner Onboarding
        IBM
        : Open Science Prize 2022
        : Educator Summit
        The Coding School 
        : Data Science TA
        : Quantum High School Research TA
        : Data Science Curriculum Development 
        
    section USC | 2023 - 2025
        DSCI 556 : User Experience Design TA
	section The Now
		[Your Company Here]
		: Data Scientist
		

Link to original

Catch me at the 🧗🏻‍♂️gym coaching, out at a ☕️ shop, or making 🎨

HCI

Human-Centered Socially Assistive Robots in ADHD/ASD: HCI Capabilities Blueprint

• In this work, we explore prior research on clinical studies for Attention-deficit/hyperactivity disorder (ADHD) and Austim Spectrum Disorder(ASD) to identify evidence-based psychosocial interventions to inform the integration of Socially assistive Robotics (SAR) into technology-based therapeutic interventions in two subtypes of Neurodevelopmental Disorder (NDD) populations: comorbid ASD and ADHD and college students with ADHD.

Preprint | Presentation

HRI mentalhealth humanfactors

Link to original

Conversational Derailment in Dispute Resolution

• This repository contains a tailored implementation of the Conversational Recurrent Architecture for ForecasTing (CRAFT) neural model, originally introduced in the EMNLP 2019 paper Trouble on the Horizon: Forecasting the Derailment of Online Conversations as they Develop to the dispute resolution domain. We pre-train the CRAFT model architecture with a custom corpus of CaSiNo, Deal no Deal, and KODIS dialogs and fine-tune on the KODIS dataset to research whether we can learn unsupervised representation of conversational dynamics in negotiation-based dialogues and exploit the structure via supervised learning for predicting for outcomes in Dispute resolution (KODIS).

Github Project

PyTorch NLP MLFlow RayTune

Link to original

Data Science

MIT IQHack23 2nd Place Project: Quantum Music Generator

• We encode sheet music into a quantum circuit and use a variational algorithm (parameter-shift rule) to create a jazz variation of the randomly generated sheet music. We extract the MFCCs of our music data and use gradient descent to optimize the hyper-parameters of the quantum gates representing the pitch and length of notes, testing similarity of our sheet music MFCC to our baseline Jazz music from the GZSTAN dataset via KNN.
• This project is the second place winner of MIT IQuack2023.

Github

quantum PyTorch optimization

Link to original

CRAFT NLP Model: Evaluating Utterance Exposure and Imbalance Strategies on fine-tuning performance for derailment prediction

• This project analyzes the model sensitivity to utterance variations and imbalance handling on 9 model variants. We determine whether CRAFT is a good predictor of conversation-level derailment on KODIS using Conversations Gone Awry Dataset (CGA-WIKI)as a baseline comparison. We compares variants using standard classification metrics (F1, AUC, Calibration Curves) at each model’s Youden-optimized threshold on the same Ground test set (CGA-WIKI), with additional diagnostics (horizons, distributions, frustration correlation, token analysis) to interpret differences.
• Including the submit agreement utterance makes KODIS fine-tuned models more prone to forecasting similar derailment scores for all conversations leading to missed conversation dynamics

Presentation | Github

PyTorch NLP explainability

Link to original

Stable Decision Tree Method for Predicting Suicidal Ideation for At-Risk Homeless Youth

• This project implements the stable decision tree algorithm based on the method outline in Improving Stability in Decision Tree Models that presents a unique distace metric for heuritic-based decision trees as a measure of stability. The algorithm produces a Pareto optimal set from which a single final optimal tree is selected according to an objective function targeting a unique metric to optimize (AUC, distance, combined, etc.).
• Improved suicidal ideation prediction accuracy by 15% with respect to the original implementation  among youth experiencing homelessness.

Presentation | Github

decisiontrees stability optimization

Link to original

Notion AI: RAG Reasoning and Prompt Case Study

• This case study designs an LLM ranking methodology for Notion-augmented RAG chatbot responses. Using A/B testing on 5 transcripts (each with two model responses), an LLM first generates a structured summary of the RAG retrieval trace guided by a five-dimension rubric (Clear, Relevant, Accurate, Functional, Transparent/Tone). We then programmatically validate summary faithfulness against the original transcript using a deterministic regex parser. A corrective loop re-queries the LLM with error evidence until all faithfulness checks pass. Perception quality of the LLM responses is measured separately via a Likert scale against the rubric.
• A second study distills prompt-level interventions to reduced hallucination rate and promot relevance runs without changing the model

Presentation

llm uxui RAG explainability

Link to original

Multivariate Time Series Classification - Activity Data from Wireless Sensor Network

• This project tackles human-activity recognition on the UCI AReM wireless-sensor dataset using a classical multivariate time-series pipeline. Time-domain features are engineered per channel and within-instance dynamics are captured by segmenting each series into ℓ chunks. Models compare (i) binary bending vs. non-bending using p-value-guided RFE + logistic regression vs. L1-penalized LR, and (ii) multiclass activity recognition using multinomial LR (L1) and Naive Bayes, with a Gaussian NB + PCA variant to decorrelate features. Selection and evaluation use nested/outer CV with ROC/AUC, accuracy, and confusion matrices.
• Adding PCA to a Gaussian NB pipeline improved multiclass CV test accuracy to .84

Presentation | Github

timeseries naivebayes scikit-learn logisticregression

Link to original

Concatenanted Dynamic Decoupling

• This notebook simulates concatenated dynamical decoupling (CDD) on a qubit coupled to a finite bath, and evaluates robustness via Uhlmann fidelity as a function of both time and concatenation level m (up to $m=20$). The method numerically evolves the state under a system–bath Hamiltonian with inserted CDD pulse sequences, then computes fidelity curves to quantify error suppression.
• ==Results show a clear optimal sequence depth around $m\approx 6$ for a fixed time $t$—beyond that, performance begins to fall off (over-concatenation overhead outweighs gains)==

Presentation | Github

quantum mathematica errorcorrection optimization

Link to original

Code

Pauli Operator Package

• lightweight Python toolkit for experimenting with Pauli matrices and the Pauli algebra: it generates and validates random Hermitian matrices, manipulates single-qubit Pauli elements with exact phase tracking, scales to multi-qubit tensor products with symbolic addition/multiplication, and decomposes matrices into the Pauli basis so you can prototype quantum-computation identities directly in code.

Github

python quantum

Link to original

Product

Solar Panel Roof Project Plan

• This report proposes a 1 MW solar roof installation on a 50,000 sq ft USC facility, framed as a full project plan including scope, WBS/RAM , risk analysis, cost/schedule, and close-out. It targets Assignment:Earth goals while detailing permitting, safety, and stakeholder reviews. Financially, the plan estimates a $5.38M project cost; with a 15% markup and a 30% tax credit, USC’s net price is ~$4.33M, backed by a contingency approach and profit sensitivity notes. Risk is quantified via a probability–consequence model yielding RF = 0.557 (medium), and the leveled schedule indicates a finish of Feb 17, 2026.
• Product takeaway: a de-risked, budgeted, and scheduled path to on-campus clean energy with explicit governance gates and measurable acceptance tests. success criteria: ≥85% performance, 2-year max duration

Presentation | Report

riskmitigation government sustainability

Link to original

Mishka's Pawsitive Pet Care

Scaled pet-care business to include B2B partnerships across two tracks—Concierge Care Packages for residential communities, and Service Referral Programs for pet-adjacent businesses to drive acquisition and retention with clear, repeatable offers working. Set up Square, Facebook ad-campaigns, and Instagram Content Drove sales by 27% by end of summer

Website

B2B brandstrategy dogwalking webdesign

Link to original

Design

Vector Logos Portfolio

• Portfolio of various logo designs for clients created with Adobe XD and brand-systems thinking processes. Clients include Twitch streamers and student organizations.

Instagram

logo adobeXD brandstrategy

Link to original

S.Q.U.I.D. 2023

• Created logo, website page, social media content, and brand strategy for global undergraduate research conference event.

Website | Instagram

brandstrategy logo UXUI webdesign

Link to original

UT Quantum Collective Brand Strategy

Created logo design, social media assets, and website with Canvas and Webflow

Website| Instagram

webflow canva logo adobeXD

Link to original

DSCI556 - Figma Tutorials

Created high fidelity web mock up tutorial with setting up frames and pages, establishing color/text styles, using Auto Layout, building components with variants, and wiring simple prototypes (micro-interactions, flows).

Figma Presentation

figma UXUI webdesign

Link to original

Publications

A Survey of Methods for Mitigating Barren Plateaus in Parameterized Quantum Circuits

• Barren Plateaus are a formidable challenge for hybrid quantum-classical algorithms that lead to flat plateaus in the loss function landscape making it difficult to take advantage of the expressive power of parameterized quantum circuits with gradient-based methods. Like in classical neural network models, parameterized quantum circuits suffer the same vanishing gradient issue due to large parameter spaces with non-convex landscapes. In this review, we present an overview of the different genesis for barren plateaus, mathematical formalisms of common themes around barren plateaus, and dives into gradients. The central objective is to provide a conceptual perspective between classical and quantum interpretations of vanishing gradients as well as dive into techniques involving cost functions, entanglement, and initialization strategies to mitigate barren plateaus. Addressing barren plateaus paves the way towards feasibility of many classically intractable applications for quantum simulation, optimization, chemistry, and quantum machine learning.==

Publication

quantum optimization

Link to original

Awards

Awards List

• USC Annenberg Fellowship

Link to original

Public Service

Activities and Organizations

• IBM Educator Summit 2022
  • Guest Speaker, promoted student-lead educational initiatives
• The Coding School
• Teaching assistant, quantum and data science programs
• UT Quantum Collective
  • Ecosystem of undergraduate initiatives for quantum computing
• Quantum Coalition
  • Led creator, S.Q.U.I.D. 2023 global undergraduate research

Link to original

Find Me

Socials

• Linkedin
• Github
• Google Scholar

Link to original