Antonio Neto

About

I am a final-year Computer Science undergraduate at PUC Minas, Brazil, graduating in December 2026, and an AI engineer at Deeptera.

I have five peer-reviewed papers at the Brazilian conferences SBPO, SIMPEP and SBAI, four as first author. They cover metaheuristics for routing and scheduling, reinforcement learning for dynamic pickup and delivery, and LLM-based multi-agent systems for industrial software.

Research interests

• Combinatorial optimization
• Metaheuristics (genetic algorithms, VNS)
• Learning to optimize and reinforcement learning for routing
• LLM-based multi-agent systems for industrial decision support

Publications

Peer-reviewed papers at Brazilian national conferences. The PDFs are the authors' versions and are written in Portuguese; each entry gives an English title and summary.

1. An Attention-Based Approach to the Dynamic Pickup and Delivery Problem

   Original title: Uma Abordagem Baseada em Atenção para o Problema de Coletas e Entregas Dinâmicas

   Antonio Neto, Rodrigo Drummond Macedo, Arley Henrique Duque, Sandro Jerônimo de Almeida

   LVIII SBPO (Brazilian Symposium on Operations Research), 2026 · Accepted

   An attention model trained with reinforcement learning and combined with VNS local search. On the ICAPS 2021 benchmark, average cost is 16.3% below Tabu Search (global average over instance groups 1–64).

   PDF (author's version, in Portuguese)

2. Optimization of Bulk Carrier Loading and Unloading Operation Sequencing with Trim Constraints Using a Genetic Algorithm

   Original title: Otimização do Sequenciamento de Operações de Carregamento e Descarregamento de Navios Graneleiros com Restrição de Trim via Algoritmo Genético

   Antonio Neto, Rodrigo Drummond Macedo, Carolina S. Almeida, Claudio Roberto Xavier, Tereza Cristina Vieira dos Santos, Marcelo Rocha Cardozo, Osmar Selhorst Filho, Mariana Coelho de Frias, Maycon Douglas Fernandes Lira, Sandro Jerônimo de Almeida

   LVIII SBPO (Brazilian Symposium on Operations Research), 2026 · Accepted

   On 112 real operational plans from the logistics operator VLI, trim-reference compliance rose from 29.12% to 84.80% and average operation time fell by 3.03%.

   PDF (author's version, in Portuguese)

3. Multi-Period Drone Routing with Time Windows and Gradual Deployment of Charging Stations

   Original title: Roteamento Multi-Período de Drones com Janelas de Tempo e Disponibilização Gradual de Estações de Recarga

   Antonio S. C. Neto, Zenilton K. G. do Patrocínio

   XXXIII SIMPEP (Brazilian Symposium on Production Engineering), 2026 · Accepted

   Undergraduate thesis. MP-EAVNS is an energy-aware Variable Neighborhood Search for a multi-period E-VRPTW. Over 9,936 runs on 92 benchmark instances, every solution was feasible with all stations available (98.6% under gradual deployment), and routes were 6.7% shorter than an adapted ALNS.

   PDF (author's version, in Portuguese) · Code

4. DeepFactory: A Multi-Agent System for Industrial Production Management and Optimization

   Original title: DeepFactory: Sistema Multiagentes para Gestão e Otimização de Produção Industrial

   Rodrigo Drummond Macedo, Antônio Neto, Arley Henrique Duque, Sandro Jerônimo de Almeida

   LVIII SBPO (Brazilian Symposium on Operations Research), 2026 · Accepted

   A prototype multi-agent platform that integrates material planning, MIP-based scheduling, predictive maintenance and computer-vision quality inspection, illustrated on a simulated pharmaceutical manufacturing scenario.

   PDF (author's version, in Portuguese)

5. An LLM-Based Multi-Agent System Integrated into Ship Loading and Unloading Management and Optimization Software

   Original title: Proposta de um Sistema Multiagentes baseado em LLM: Integração em um Software de Gerenciamento e Otimização de Carregamento e Descarregamento de Navios

   Antonio S. Neto, Rodrigo D. Macedo, Carolina S. Almeida, Arley Henrique Duque, Sandro Jerônimo de Almeida

   SBAI-SBSE (Brazilian Symposium on Intelligent Automation), 2025 · Published

   Four specialised LLM agents (Manager, Helper, Optimizer and Data Analytics) integrated into conventional ship-operation software to register ships, generate optimized plans and build visual analyses.

   PDF (author's version, in Portuguese)

Education

• B.Sc. in Computer Science, Pontifical Catholic University of Minas Gerais (PUC Minas), Brazil2023 – Dec 2026 (expected)

  Thesis: multi-period drone routing with time windows and gradual deployment of charging stations, advised by Prof. Zenilton K. G. do Patrocínio. Selected grades (out of 100, pass mark 60): Thesis Project I 98, Design and Analysis of Algorithms 88, Artificial Intelligence 82, Graph Theory and Computability 80.

• Mandarin, Confucius Institute at UFMG2024 – present

  Currently at HSK 4 level.

Experience

• Artificial Intelligence Engineer, DeepteraApr 2026 – present

  Part-time R&D on optimization and multi-agent systems. This work and my thesis led to four papers accepted at SBPO and SIMPEP 2026.

• Computing and Data Intern, DeepteraAug 2024 – Apr 2026

  Genetic-algorithm optimization and backend services for the logistics operator VLI; first author of an LLM multi-agent paper published at SBAI-SBSE 2025.

Selected projects

• Drone routing (thesis code)

  Reproducible GRASP/VNS experiments for multi-period drone routing on E-VRPTW benchmark instances.

• Anki MCP

  An MCP server that lets AI assistants create and review Anki flashcards; I use it for my Mandarin study.

Skills and languages

Programming languages

Python, C, TypeScript

Agent systems

LangChain, LangGraph, RAG, MCP

Machine learning

PyTorch, scikit-learn

Data science

Pandas, NumPy and data visualization

Backend

Django, FastAPI, REST APIs

Databases

PostgreSQL, MySQL

DevOps

Azure DevOps, Docker

Version control

Git, GitHub

Optimization

Mathematical modeling, solvers and metaheuristics

Cloud

Azure, Google Cloud

Languages

Portuguese (native), English (fluent), Mandarin (HSK 4)

Contact

antonio.couto@sga.pucminas.br

antonionetodev@yahoo.com

github.com/nietus · LinkedIn