CV
Download: CV (PDF)
Summary
- Generative AI Scientist with a PhD in AI, specializing in AI systems that adapt to real-world change. Built first-of-its-kind AI agents for production model monitoring and improvement (CAMA at AAMAS, KC-Agent at IEEE COMPSAC).
- Currently building healthcare cognitive architectures at Cotiviti (Cotiviti press). Also contributed to diabetic retinopathy research at Emory Medical Center. 7+ years of research at Barcelona Supercomputing Center/Lenovo developing continual learning frameworks for drift datasets (Lenovo press).
- Published in top-tier AI venues (AAMAS, ICPR, VISAPP). PhD research achieved 99% storage reduction while maintaining 96.9% accuracy in privacy-sensitive healthcare applications through autonomous prompt evolution (Springer Nature).
- Some facts: I prototype fast. I measure everything (costs, latency, performance). I follow the old-school unix principles: do one thing well.
Stack Software
Linux, Python, PyTorch, Numba, LangGraph, AutoGen, Langchain, transformers, MLflow, Ray, ChromaDB, Kubernetes, Node.js, Spark, LLMs
Professional Experience
Generative AI Scientist | Cotiviti
Apr 2026 - Present | New York, USA
- Concept extraction for Summary plan description documents using connected components (graph), RAG and anomaly detection. Really cheap pipeline: only $10 for 1k docs end-to-end with $3M-$5M annual revenue opportunity.
- Building a cognitive architecture that enables agents to learn fast using semantic, episodic and procedural memories. Part of this work is already used in internal projects.
Generative AI Research Engineer (Intern) | Cotiviti
Jul 2024 - Present | New York, USA
- Built a 3D RAG system (semantic, lexical and graph) with a cost-efficient pipeline for parsing and retrieving complex documents (1.5k docs for $0.5). All open source and fast execution.
- Built an evaluation workflow for benchmarking RAG architectures using the MMLU methodology (multi-choice options).
Recognised Researcher | Barcelona Supercomputing Center & Lenovo
Feb 2026 - Present | Spain/USA, part-time
- Research on cognitive architectures for agents. Just published KC-Agent at COMPSAC 2026.
AI Research Engineer | Barcelona Supercomputing Center & Lenovo
Feb 2019 - Jan 2026 | Spain/USA
- Architected cognitive frameworks for Language Agents that autonomously monitor ML model health, reducing degradation incidents and improving model reliability
- Built the TADIL algorithm for domain-agnostic continual learning, enabling 6% performance improvement across diverse domains without requiring task labels
- Developed Scanflow, an MLOps platform for debugging ML workflows; several projects were built on top of it
- Developed a novel drift detection algorithm using autoencoders for drift detection on images, published in a top-tier journal
- Led cross-functional teams, resulting in several publications in top-tier journals and conferences
AI Engineer Intern | Emory Global Diabetes Research Center
Jul 2023 - Jul 2024 | USA
- Developed a resource-efficient cognitive architecture for LLM-agents using finetuned small language models (Llama, Phi), reducing API costs by 35% while maintaining reasoning capabilities
- Built a privacy-preserving domain adaptation algorithm for diabetic retinopathy detection, achieving 96.9% accuracy with 99% storage reduction across diverse patient populations
- Published 3 papers on continual learning for medical imaging, including a Best Paper Nominee at VISAPP 2025
AI Consultant | SUNAT - Inter-American Development Bank
Jan 2022 - Dec 2025 | Peru
- Built the first agent-based system for customs to monitor ML models in production, enabling real-time detection of customs fraud patterns
- Built the first “Improver” agent for automated ML model updates, reducing manual intervention by 65% while maintaining model performance
Data Scientist | SUNAT (Tax Administration)
Nov 2017 - Nov 2018 | Peru
- Engineered an ensemble-based fraud classifier with 92% accuracy for identifying unjustified patrimony
- Developed graph analysis methods to uncover complex tax evasion networks, processing 1TB+ of financial data monthly
- Built scalable analytical infrastructure processing millions of electronic invoices, directly contributing to a $10M increase in tax revenue
Education
- PhD in Artificial Intelligence, UPC BarcelonaTech, Spain (2021-2025)
- Research focus: “Human-on-the-loop Continual Learning: Data, Knowledge and Agents for Model Adaptation”
- Highest honors: Cum laude (Thesis: UPC repository - TDX)
- MSc in Artificial Intelligence, UPC BarcelonaTech, Spain (2019-2021)
- Thesis: “Scanflow: A Learning-symbolic Framework for ML Workflow Debugging”
- Published in Expert Systems With Applications (Impact Factor: 8.665)
- BSc in Computer Science, Magna Cum Laude, National University of Engineering, Peru (2011-2016)
- Top 5% of graduating class
Publications
2026
- KC-Agent: A Dual-Process Cognitive Architecture for Efficient ML Model Improvement. IEEE COMPSAC 2026. arXiv
- Adaptive Prompt Evolution for Continual Learning in Diabetic Retinopathy Detection. SN Computer Science (Springer Nature). Link
2025
- PeruMedQA: Benchmarking Large Language Models (LLMs) on Peruvian Medical Exams - Dataset Construction and Evaluation. arXiv. Link
- Bravo-Rocca, G., et al. “Feature Engineering for Agents: An Adaptive Cognitive Architecture for Interpretable ML Monitoring.” AAMAS 2025. ACM DL - arXiv
- Bravo-Rocca, G., et al. “Experience Replay and Zero-shot Clustering for Continual Learning in Diabetic Retinopathy Detection.” VISAPP 2025. Porto, Portugal. [Best Student Paper Award Finalist] DOI
2024
- Bravo-Rocca, G., et al. “TADIL: Task-Agnostic Domain-Incremental Learning through Task-ID Inference using Transformer Nearest-Centroid Embeddings.” ICPR 2024. Kolkata, India. [Oral Presentation] Springer - arXiv
2022
- Bravo-Rocca, G., et al. “Scanflow: Multi-graph framework for Machine Learning workflow management, supervision, and debugging.” Expert Systems With Applications. [Impact Factor: 8.665] ACM DL
- Scanflow-K8s: Agent-based Framework for Autonomic Management and Supervision of ML Workflows in Kubernetes Clusters. IEEE CCGrid 2022. DOI
- Human-in-the-loop online multi-agent approach to increase trustworthiness in ML models through trust scores and data augmentation. IEEE COMPSAC 2022. DOI
2021 and earlier
- Scanflow: an end-to-end agent-based autonomic ML workflow manager for clusters. ACM Middleware 2021 (Demo/Poster). DOI
- Energy-Aware Dynamic Pricing Model for Cloud Environments. GECON 2019. DOI
- Sparkmach: A Distributed Data Processing System Based on Automated Machine Learning for Big Data. SIMBig 2018. DOI
- Bluetooth-based indoor localization mechanisms. Sensors 2017 (two papers); Citizen security using machine learning algorithms through open data. LATINCOM 2016
Full list on DBLP
Awards and Recognition
- PhD highest honors (2025) - Cum laude for work on Continual Learning and Agents
- Intel-Lenovo AI Innovators Award (2023) - Selected for innovation in continual learning and cognitive architectures research
- María Rostworowski I Researcher (2020) - National recognition for excellence in AI research
Key Projects
- Cognitive Architecture for ML Monitoring (CAMA) (github.com/gusseppe/cognitive_architecture_checker)
- LLM-based agent framework that autonomously monitors ML models in production environments
- Novel decision procedure algorithm and structured memories for interpretable model diagnosis
- Featured in AAMAS 2025 conference; developed in collaboration with Lenovo AI Lab
- Successor: KC-Agent (arXiv), IEEE COMPSAC 2026
- TADIL / TADILER (github.com/gusseppe/TADIL - github.com/gusseppe/tadiler)
- Scanflow (github.com/gusseppe/scanflow)
- Production-grade MLOps framework for ML workflow management, debugging, and reliability
- Many projects were inspired by this framework (Scanflow-k8s, CAMA, KC-Agent)
- Published in top-tier journal “Expert Systems With Applications” (Impact Factor: 8.665)