Cindy Eliana Levi
About
I'm a computer scientist from the University of Buenos Aires working on computational mental health. I study speech and language as windows into psychological states, and I'm interested in how to design and evaluate AI systems that affect people's well-being. I'm also drawn to the computational analysis of literary texts.
I'm a volunteer research assistant at the ARC Lab (Child Mind Institute), advised by Daniel M. Low, PhD, where I'm co-writing a methods manuscript on open-source tools for coding psychological constructs in text with large language models. With him I'm also building a model to support suicide-risk triage from clinical records in the Province of Jujuy, Argentina. My M.Sc. thesis at UBA, advised by Sebastián Uchitel and Hernán Gagliardi, was on compositional controller synthesis and was presented at FAACS 2026.
Computer scientist & researcher.
- Email: cindylevi@gmail.com
- City: Buenos Aires, Argentina
- Degree: M.Sc. in Computer Science (UBA)
- GitHub: github.com/cindylevi
I also teach Operating Systems at UBA as a teaching assistant. Alongside research, I work as a Senior Backend Engineer at Brubank, a digital bank, building services and data pipelines that handle sensitive financial data; before that I was a backend engineer at Modak and Mercado Libre, starting in 2021.
Outside research, I've been part of Chicas en Tecnología since 2017, and in 2026 I represented UBA at the National Model United Nations in New York, where I received an Outstanding Delegate award in the NPT Committee together with a fellow delegate.
I'm applying to PhD and research master's programs for Fall 2027.
Research
My interests lie at the intersection of computation and the social and psychological sciences: computational mental health, speech and language as windows into psychological states, and the design and evaluation of AI systems that affect people's well-being. I am particularly drawn to work that pairs methodological rigor (formal reasoning, machine learning, statistical modeling) with real-world social impact, and to research questions whose answers change how people are cared for. I am also interested in the computational analysis of literary texts.
ARC Lab (AI, Risk and Contemplative Science), Child Mind Institute
2026 - Present
Volunteer Research Assistant · Advisor: Daniel M. Low, PhD
- The lab studies mental health through computational analysis of speech and language data.
- Co-writing a methods manuscript introducing clinicians to the lab's open-source tools llm_tracker (LLM-based coding of psychological constructs in text) and anonymize-pii (de-identification).
- Wrote two tutorials, merged into llm_tracker: de-identifying clinical text before LLM coding, and mapping LLM-coded quotes with sentence embeddings to check whether constructs are separable.
Suicide Risk Triage Support Model, Province of Jujuy, Argentina
Jun 2026 - Present
Research project · with Daniel M. Low, PhD
- Built an interpretable model that suggests the suicide-risk level (low/medium/high) clinicians assign under a provincial protocol, from 1,468 labeled episodes, with a leakage audit and patient-grouped splits.
- Compared 8 classifiers (macro-F1 ≈ 0.45 vs. 0.33 chance); a cost-sensitive decision rule cut severe misses (high risk predicted as low) from 24 to 0; audited performance by sex and region.
- Coded clinicians' free-text notes against a Spanish adaptation of the Suicide Risk Lexicon with llm_tracker.
M.Sc. Thesis: Scaling Compositional Controller Synthesis with a New Minimization Technique for Branching Bisimilarity
2024 - 2026
University of Buenos Aires (UBA) · Advisors: Sebastián Uchitel and Hernán Gagliardi
- Replaced the bottleneck minimization step of compositional controller synthesis (WSOE) with branching bisimilarity, proving the conditions under which the replacement is sound.
- Implemented it in MTSA and validated it on 2,212 real instances: under 1,000 ms on the largest ones (vs. 7,600 ms for WSOE), with the same reduction in 83% of cases.
Publications & Presentations
Manuscripts in preparation
LLM Tracker: Open tools for coding psychological constructs in text with large language models
C. E. Levi and D. M. Low. Methods review for clinical and psychological researchers. In preparation.
Conference presentations
Scaling compositional controller synthesis with a new minimization technique for branching bisimilarity
C. E. Levi, H. Gagliardi, and S. Uchitel. Talk at FAACS 2026, Taller Argentino de Fundamentos para el Análisis y la Construcción Automática de Software, Villa Carlos Paz, Córdoba, Argentina, Mar. 2026.
Open-source research software
Tutorials for llm_tracker
C. E. Levi. “Removing personal information before LLM coding” and “From coded quotes to a map,” contributed to llm_tracker, Child Mind Institute, 2026. github.com/childmindresearch/llm_tracker
Experience
Education
M.Sc. in Computer Science
2025 - Aug 2026
University of Buenos Aires (UBA), Faculty of Exact and Natural Sciences
Thesis advised by Sebastián Uchitel and Hernán Gagliardi (see Research).
B.Sc. in Computer Science
2021 - Dec 2024
University of Buenos Aires (UBA), Faculty of Exact and Natural Sciences
GPA: 8.23 / 10.00 (3.29 / 4.00 converted), above program average.
Awards & Service
Outstanding Delegate(s), NPT Committee — NMUN New York
Apr 2026
Represented the Syrian Arab Republic for the University of Buenos Aires; awarded jointly with a fellow delegate.
Speaker — JIS Go Live, Hospital Italiano de Buenos Aires
2020
Presented Iuvo, an app designed to improve healthcare services for older adults.
Chicas en Tecnología — PUMM Program
2017 - Present
Member of a community that promotes access to technology for young women in Latin America.
Industry Experience
Senior Backend Engineer
Sep 2025 - Present
Brubank (digital bank)
Build services and data pipelines that handle sensitive financial data under regulatory constraints.
Backend Engineer
Dec 2023 - Sep 2025
Modak (fintech startup)
Led projects end to end and built the company's AWS cloud infrastructure from scratch.
Backend Developer
Mar 2021 - Dec 2023
Mercado Libre
Built high-throughput services, image-processing pipelines and internal libraries.
Skills
Methods
Natural language processing, LLM-based text coding, machine learning, statistical analysis, formal methods (controller synthesis, behavioural equivalences)
Research tools
Python (pandas, NumPy, scikit-learn, PyTorch, Hugging Face, spaCy, OpenAI and Anthropic APIs), R, SQL, LaTeX, Git
Engineering
Go, Java; distributed backend systems, data pipelines, AWS, PostgreSQL
Languages
Spanish (native); English (advanced, Cambridge First Certificate; IELTS in progress); Hebrew (basic)
Teaching
Graduate Teaching Assistant, Operating Systems
Mar 2025 - Present
University of Buenos Aires (UBA), B.Sc. in Computer Science
Teach lab sessions, design and grade assignments, and mentor undergraduate students.
Projects
Selected course projects
Predicting tumor prognosis from gene expression: a comparison of classical ML models
Compared decision trees, KNN, LDA, naive Bayes, SVMs and random forests on RNA measurements of 200 genes from 500 pre-tumoral samples with imbalanced classes; an RBF-kernel SVM reached 0.92 AUC-ROC on held-out data.
Reward hacking: origins, theory and mitigation
Foundations of AI, UBA, 2026
A review of reward hacking from early rule-based systems to RLHF-trained language models, its roots in Goodhart's law, and current mitigation strategies.
Spectral analysis of social networks
With M. Zeitoune and A. Azar · Numerical Methods, UBA, 2023
Implemented the power method with deflation and used eigenvector centrality and Laplacian eigenvectors to recover the community split in Zachary's karate club (r = 0.81) and to relate user attributes to friendships in the ego-Facebook network.
- All
- Engineering
- Social Impact
Iuvo
Iuvo, an app to improve the health system for the third age
Contact
Location
Buenos Aires, Argentina
Call
+54 9 11 3012-3847