Computer Science senior (GPA 9.23/10) · Python back-end engineer and applied AI/security researcher.

I build secure, high-throughput back-end systems and do applied AI and security research: data pipelines, LLM/SLM systems, and statistical modeling for Petrobras (with CeMEAI/USP), RNP, and ITA, plus current adversarial-AI security research at Poli-USP/LARC (Bradesco project). 19 accepted peer-reviewed publications (10 first-author), including BRACIS 2026 (Springer LNAI) and the SBRC 2026 Best Artifact Award. ICPC Silver Medalist. Most of my papers ship with a public reproducibility artifact.

• Pre-Master's Research Fellow @ USP (Poli-USP/LARC), AI Security: adversarial-AI / GenAI security in a Bradesco project, red team and blue team of LLMs (prompt injection, jailbreaks) and defenses (guardrails, input/output filters, detection).
• R&D Fellow @ ITA (CYBERGUARD), CNPq (Apr–Jul 2026): co-authored the SBSeg 2026 tutorial Inteligência Artificial aplicada à Resposta a Incidentes, contributing the case studies on ticket anonymization, incident classification, and LLM-assisted response playbooks.
• Back-end Engineer @ Alice Humam (Petrobras, with CeMEAI/USP, 2025–2026): built the Python/FastAPI back-end of a reliability-analysis platform: ETL pipelines and survival-analysis models translated from research R code and verified numerically equivalent, plus the HTML reporting engine and a per-route observability layer (p50/p95/p99); over 16 months, 111k+ lines and 375 merged MRs, became the team code reviewer, and co-managed the dev team (Scrum/Kanban).
• Competitive Programming: Silver Medal 🥈 at the 2025 ICPC Brazil Regional First Phase (team Array de Noobs 2.0).

• AnonShield: Scalable On-Premise Pseudonymization (First author · SBRC 2026 Best Artifact, all 4 reproducibility badges) High-throughput pseudonymization for CSIRT vulnerability data. GPU-accelerated NER, streaming I/O, and LRU caching cut processing of a 550 MB dataset from over 92 hours to under 10 minutes (738× faster), at 94.2% F1 and 96.4% recall, GDPR/LGPD-compliant without losing analytical utility.
• Decomposing Memorization Reduction in Privacy-Preserving Fine-Tuning of SLMs (First author · BRACIS 2026, Springer) First empirical study of how DP-SGD and HMAC pseudonymization compose when fine-tuning 1 to 3B language models on CSIRT data, across 96 LoRA adapters and a dual extraction attack.
• Not All 4-bit Quantizers Are Equal (First author · SBSeg 2026, full paper) The choice of 4-bit quantizer changes how much PII a fine-tuned model leaks at deployment time, not just its speed. AWQ, GPTQ and GGUF compared across five 0.5–7B SLMs, with the calibration-corpus axis isolated by four ablations. Artifact reproduces 141/141 published values.
• Large-Scale Docker Hub Security Measurement (First author · two SBSeg 2026 short papers) Distributed Go crawler over 12M+ repositories, an 84M-node image dependency graph in Neo4j, and 50k+ high-exposure images scanned with six scanners into a public 170M-finding dataset. Two artifacts: a multi-scanner census of Linux base images and a uniform random-sample measurement of the registry.
• ZeroLINC: Training-Free Incident Triage (First author · SBSeg 2026, tool track) Classifies CSIRT/SOC incident reports into the 12 NIST SP 800-61r3-derived categories with no training: a k-NN instance-memory engine at 90.8% and a zero-shot engine, behind one command and running locally. Measurement study behind it: 293 runs.
• CryptoCensus: Post-Quantum Readiness of Docker Hub (First author · WTICG/SBSeg 2026) Distributed census of 11,962 images and over 4.2 million cryptographic assets, measuring how far the registry is from post-quantum migration.

More reproducibility artifacts: RAGtrap (signed provenance with O(1) traceback for poisoned RAG corpora) · PixGuard-Sim (deadline-aware testbed for Pix fraud detectors) · wazuh-study (LLM-generated SIEM rules on real SSH logs) · AdminForge (agentless privileged-identity management for Linux fleets) · MulitaMiner (LLM extraction of OpenVAS reports, 93.55% F1).

Languages

Back-end

Data Science & ML

AI · LLMs · NLP

Front-end & Mobile

Databases

DevOps & Observability

Security & Crypto

Also hands-on with: Loki, Alertmanager, Grafana Alloy, Uptime Kuma, GraphSAGE, WeasyPrint, sentence-transformers (e5, bge-m3), BM25, BERTScore, Qwen-VL, PaddleOCR-VL, EMV/BR Code parsing, and AI-assisted development (Claude Code, OpenAI Codex, Gemini CLI).

• 🥇 Best Artifact Award, SBRC 2026 (best artifact of the entire conference) and Distinguished Artifact Reviewer, SBRC 2026
• 🥈 Silver Medal, ICPC/SBC Programming Marathon 2025 (Brazil Regional First Phase); 3rd place RS, Phase Zero
• 🏆 1st Place, SBRC 2026 Hackathon · 2nd Best Paper, WRSeg/ERRC 2025
• 🎓 Hackers do Bem (144h, MCTI/SENAI/RNP) · ICT Residency in AI & Data Science (180h, BRISA/Softex) · CS50P (Harvard)

📫 Let's connect: LinkedIn · Lattes · Codeforces · Email