Esteban Schafir
AI Engineer / Data Engineer
Applied AI Researcher • Specializing in LLM Agents & Data Systems
I design and build agentic AI systems, scalable data engineering pipelines, and complex reasoning frameworks where accuracy, verification, and data reliability are foundational requirements. Combining 4+ years of academic AI research at FIU with 10 years of enterprise IT & security engineering.
Bridging Applied AI, Data Reliability & Systems Security
A high-rigor engineering mindset honed through 4+ years of academic research and 7 years in enterprise infrastructure.
Background & Expertise
I am an AI Engineer, Data Engineer, and Ph.D. candidate in Computer Science at Florida International University (FIU). My work spans large language models, multi-agent systems, retrieval-augmented generation (RAG), and end-to-end data pipelines.
Before diving deep into academic AI research, I spent over 10 years in enterprise IT, information security, and risk engineering at organizations like Equifax and Deloitte partners. That background shapes how I build AI today: not as brittle demo toys, but as hardened, deterministic systems with strict guardrails, transactional rollbacks, and measurable performance.
At FIU, I conduct research on hierarchical decomposition for Text-to-SQL, visual document retrieval via teacher-student model distillation, and graph neural network robustness, publishing in premier venues including AAAI, IEEE TPAMI, and the ACM Web Conference.
Core Specialization Domains
Agentic AI & LLMs
Architecting multi-agent systems (LangGraph), DAG-based problem decomposition, and constrained generation with deterministic verification loops to prevent hallucinations.
Data Pipelines & SQL
Designing end-to-end ETL/ELT pipelines in Python and SQL. Integrating relational databases, vector embeddings, and building natural-language Text-to-SQL query compilers.
Empirical AI Research
Conducting Ph.D. research at FIU on sub-10ms visual document retrieval via teacher-student distillation, graph representation learning, and complexity-aware query routing.
Cybersecurity Foundation
10 years of enterprise risk modeling, vulnerability remediation, and ISO 27001 / NIST compliance at Equifax and Deloitte partners—ensuring data systems are secure by design.
Featured Engineering Projects
Concrete implementations spanning multi-agent reasoning, hierarchical Text-to-SQL, voice assistants, and offline-first desktop systems.
DecoSearch
Complexity-Aware Routing & Plan-Level Repair for Enterprise Text-to-SQL
A cutting-edge research framework for Text-to-SQL that decomposes complex natural-language queries into dependency DAGs, routes them based on structural complexity, and refines query candidates using Monte Carlo Tree Search and database execution feedback.
Resume Tailor AI
Multi-Agent LangGraph Pipeline with Strict Anti-Hallucination Guardrails
An autonomous multi-agent system that ingests resumes (PDF or LaTeX source) and job postings, runs gap analysis, and rewrites bullet points with an adversarial verification loop to guarantee no unsupported skills, tools, or metrics are introduced.
Recepción IA
Multi-Tenant Voice AI Assistant for Medical & Dental Clinic Phone Reception
A production SaaS platform where clinics configure autonomous Spanish-speaking phone assistants via VAPI, with live Google Calendar two-way synchronization, AES-256-GCM token encryption, Supabase RLS multi-tenancy, and turn-by-turn transcript inspection.
Lovecraft Multimodal Studio
LangGraph-Orchestrated Multimodal Suite: Lore Generation, Diffusion Art & Neural Voice Cloning
An end-to-end multimodal suite for classic literature combining stateful LangGraph orchestration, local Ollama lore generation (gemma4:e2b), int8 quantized diffusion cover art (Z-Image-Turbo), and zero-shot Vincent Price voice cloning (CosyVoice3).
Smart File Organizer
Offline-First Desktop App with an AI Proposal Engine & Self-Defending Filesystem Harness
A private local desktop application that categorizes unstructured file collections and suggests organized directory structures, executing moves and renames through a sandboxed, fully reversible filesystem harness.
Mighty Path
Autonomous DJI Drone Waypoint Mission Planner with Dynamic Terrain-Following
An ultra-premium desktop mission planner for DJI drones featuring Google Maps visual flight path plotting, dynamic Google Elevation API terrain-following, cinematic shape generators, and native DJI WPML v1.0.2 export.
TracepAI
Local-First Dockerized Financial Analytics Hub with On-Premise Receipt OCR
A private personal finance tracking platform running in Docker that parses receipts, invoices, and bank CSVs locally using Tesseract OCR, offering real-time budget analytics with zero third-party cloud data exposure.
Flowdeck
High-Performance Multi-Board Kanban Desktop App for macOS & Modern Web
A multi-board Kanban productivity desktop and web application built with Next.js, Electron, and TypeScript, featuring smooth drag-and-drop card physics, rich-text markdown notes, and local file persistence without user accounts.
AI & Computer Science Research
Peer-reviewed research conducted at Florida International University on LLM reasoning, Text-to-SQL, visual retrieval distillation, and graph neural networks.
DecoSearch: Complexity-Aware Routing and Plan-Level Repair for Text-to-SQL
E. Schafir, X. Zheng, H. A. Salehi, Z. Chen, M. Sha, W. Cheng, D. Luo
Proposes an agentic Text-to-SQL framework using hierarchical question decomposition into dependency DAGs, complexity-aware query routing, and plan-level repair with MCTS to achieve high execution accuracy on BIRD and Spider benchmarks.
The Matter of Captchas: An Analysis of a Brittle Security Feature on the Modern Web
B. Ousat, E. Schafir, M. A. Tofighi, D. C. Hoang, C. V. Nguyen, S. Arshad, A. Kharraz
Presents an extensive empirical analysis of modern CAPTCHA implementations across top web domains, identifying architectural brittleness and training CNNs/vision models to demonstrate automated bypass vectors.
Constructs of Deceit: Exploring Nuances in Modern Social Engineering Attacks
M. A. Tofighi, B. Ousat, J. Zandi, E. Schafir, A. Kharraz
Analyzes evasion strategies and multi-channel cognitive triggers in state-of-the-art credential harvesting campaigns and modern enterprise social engineering attacks.
Addressing Structural Distribution Shift in Explanations for Graph Neural Networks
Z. Chen, H. A. Salehi, E. Schafir, X. Zheng, J. Zhang, H. Wei, J. Ni, F. Shirani, D. Luo
Develops theoretical frameworks and algorithms to mitigate distribution shifts when generating faithful, robust sub-graph explanations for Graph Neural Networks (GNNs).
Explanation-Preserving Augmentation for Semi-Supervised Graph Representation Learning
Z. Chen, J. Ni, H. A. Salehi, X. Zheng, E. Schafir, F. Shirani, D. Luo
Introduces explanation-preserving data augmentations for semi-supervised graph representation learning, demonstrating superior node and graph classification performance under label scarcity.
LM2otifs: An Explainable Framework for Machine-Generated Texts Detection
X. Zheng, Z. Chen, E. Schafir, S. Chen, H. A. Salehi, H. Chen, F. Shirani, M. Sha, W. Cheng, D. Luo
Proposes a graph-perspective explainable framework to detect machine-generated text by analyzing structural motifs and syntactical dependency patterns.
In-Application Defense Against Evasive Web Scans Through Behavioral Analysis
B. Ousat, M. Shariatnasab, E. Schafir, F. S. Chaharsooghi, A. Kharraz
Presents a behavioral analysis engine embedded inside web applications to dynamically detect, delay, and neutralize evasive automated vulnerability scanners and malicious crawlers.
e-Health: An Introduction to the Challenges of Privacy and Security
H. R. Jara, E. Schafir
Early foundational research surveying electronic healthcare record vulnerabilities, regulatory compliance challenges, and cryptographic privacy primitives for patient data.
Professional & Research Experience
10+ years of combined experience across applied AI research, data engineering, and enterprise information security.
Graduate Research & Teaching Assistant
Leading applied AI research on agentic Text-to-SQL decomposition, visual document retrieval, and graph neural network robustness.
Information Security Officer
Led enterprise-level information security, vulnerability remediation pipelines, compliance programs, and executive KPI reporting across corporate infrastructure.
Information Security Analyst
Managed security monitoring, access controls, audit finding remediation, and core security infrastructure tooling for a major financial institution.
Information Security Analyst
Delivered technical vulnerability assessments, penetration testing, and security hardening audits across diverse enterprise client environments.
Categorized Technical Skills
Structured competency matrix spanning frontier LLM agent orchestration, high-throughput data engineering, and enterprise cyberdefense.
AI & Machine Learning
Deep learning models, neural graph architectures, and statistical evaluations.
LLMs & Agentic Systems
Multi-agent workflows, in-context reasoning, and deterministic verification harnesses.
Data Engineering & Databases
Scalable data ingestion, relational schema modeling, and vector search systems.
Languages & Frameworks
Production software engineering across web, desktop, and backend services.
Systems, Cloud & Tools
Containerization, cross-platform packaging, and developer automation.
Cybersecurity & Governance
Defensive engineering, compliance frameworks, and system hardening.
Education & Credentials
Advanced academic training in Computer Science and Cybersecurity from Florida International University and Universidad Nacional de Quilmes.
Doctor of Philosophy (Ph.D.)
Master of Science (M.S.)
Bachelor of Science (B.S.)
Let's Build Together
Open to AI Engineering and Data Engineering opportunities. Reach out directly via email or professional networks.
Direct Channels
Profiles & Dossier
Connect on professional platforms or download my comprehensive AI & Data Engineer resume for recruitment review.