CV

Open Innovation Engineer

📄 View Full CV (PDF)

Representation Analysis Activation Steering Large Language Models Reproducible Machine Learning Python C/C++ TypeScript JavaScript PyTorch Hugging Face Transformers Scikit-Learn SciPy NumPy Pandas Matplotlib Plotly Linux Git GitHub Google Colab Jupyter Notebook LaTeX English Spanish Representation Analysis Activation Steering Large Language Models Reproducible Machine Learning Python C/C++ TypeScript JavaScript PyTorch Hugging Face Transformers Scikit-Learn SciPy NumPy Pandas Matplotlib Plotly Linux Git GitHub Google Colab Jupyter Notebook LaTeX English Spanish

Research Interests

Representation analysis, activation steering, large language models, and reproducible machine learning.

Education

B.S. in Computer Systems Engineering
Instituto Tecnológico de Tijuana
Aug 2019 – Jan 2024
GPA: 95 / 100

Research Experience

Independent Researcher
Affective Representation Analysis — Independent Replication
Apr 2026 – Present

  • Tested whether affect-associated representation geometry and activation steering transfer to GPT-2 Medium and Gemma 4 E2B through a partial replication of selected analyses from Anthropic’s emotion-concepts study.
  • Released four reproducible Google Colab configurations spanning two models and nine- and 20-emotion label sets; extracted vectors from 2,000 generated stories and evaluated them with Logit Lens projections, PCA, cosine similarity, and steering interventions.
  • Observed partial PC1 valence-like separation, local similarity families, and model- and prompt-dependent intervention effects; documented corpus, evaluation, and control limitations in the project repository.

Undergraduate Thesis Researcher
Instituto Tecnológico de Tijuana
Aug 2023 – Feb 2024

  • Co-designed a quasi-experimental pretest-posttest study and analyzed 39 of 46 consented students: 20 in the control group and 19 in the digitally assisted group.
  • Implemented a Python pipeline for descriptive comparisons and within-group pre/post tests using Pandas, NumPy, Matplotlib, NLTK, and SciPy.
  • Found higher mean post-test scores in both groups, while the digitally assisted group did not descriptively outperform the control in this sample; documented potential confounders and avoided a causal interpretation in the thesis repository.

Research Assistant
ENLACE 2023 Summer Research Program — University of California San Diego
Jun 2023 – Aug 2023

  • Evaluated whether fully connected networks could approach a laboratory Graph Neural Network baseline for binary classification of candidate track line segments in the CMS Line Segment Tracking pipeline.
  • Co-developed and trained two fully connected PyTorch neural networks on a UC San Diego supercomputing cluster, including preprocessing, training, inference, and ROC analysis.
  • Measured a 0.9724 ROC AUC for the larger DNN on the saved held-out comparison, 0.0027 below the laboratory GNN baseline of 0.9751; coauthored the code and report and presented the result at Instituto Tecnológico de Tijuana.

Manuscript in Preparation

Partial Replication of Anthropic’s Emotion Vectors Using Gemma 4 E2B and GPT-2 Medium Inside a Google Colab T4 Notebook — 2026
A. J. Flores-Azcona. Single-author manuscript. Code and data.

Industry Experience

Open Innovation Engineer
Open Innovation — Samsung Research Tijuana
Feb 2024 – Present

  • Developed 10+ AI proof-of-concept prototypes spanning large language models, speech recognition, virtual assistants, AI agents, and on-device machine learning for the Samsung TV platform.
  • Integrated partner SDKs and APIs in C/C++, Python, TypeScript, and JavaScript to evaluate AI capabilities in resource-constrained TV environments.
  • Collaborated with international engineering teams and technology partners; synthesized prototype results and integration constraints into four technical demonstrations for executive leadership during workshops in South Korea.

Teaching and Mentoring

Academic Tutor
Universidad Autónoma de Baja California
Mar 2023 – Oct 2023

  • Tutored 62 undergraduate engineering students in Differential Calculus, Programming Methodology, and Programming and Numerical Methods through individual and group sessions.
  • Adapted problem-solving explanations to students’ prior knowledge and coordinated study groups around requested topics.
  • Analyzed attendance patterns with Pandas and Matplotlib, then proposed combined individual/group tutoring and targeted tutor training in a public report and analysis.

Selected Presentations

  • Partial Replication of Anthropic’s Emotion Vectors Using Gemma 4 E2B and GPT-2 Medium Inside a Google Colab T4 Notebook — Internal Research Seminar, Samsung Research Tijuana (May 12, 2026).
  • Tizen Native in the Age of AI Tools — INNOVATEC 2025, Instituto Tecnológico de Tijuana (Nov 20, 2025).
  • Using Machine Learning for Particle Tracking at the Large Hadron Collider — Poster at UC San Diego and bilingual oral presentation at Instituto Tecnológico de Tijuana (2023).