Lucas Couto

Machine Learning Engineer · Computer Vision Researcher

I build vision systems that run in production, and research the architectures behind them.

View work LinkedIn GitHub E-mail
Lucas Couto, seated, smiling, wearing a dark t-shirt

I'm Lucas Couto, 25, based in Brazil. I've been programming since I was 12. Today I split my time between shipping computer vision systems at a deeptech company and researching visual recommendation architectures in my Master's. Before that, I was CTO of an identity-verification startup and AI Lead at a franchising group.

13+
years programming
2
papers published · ICEIS 2025 & 2026
9
deep learning models built · 3 in production
4
engineers led as AI Lead

Experience

  1. Machine Learning Engineer · Kasco AI

    May 2026 – present · remote

    • Brazilian deeptech, a Unicamp spin-off, working in Energy, Agro and Industry.
    • Develop a cattle-counting system from drone imagery using detection and instance segmentation algorithms.
    • Work on the asynchronous video-processing pipeline built on distributed message queues.
    • Contribute to the web platform that manages field operations.
    • Wrote the Git conventions and workflow guide adopted by the team.
    • Deep Learning
    • Computer Vision
    • Message Queues
    • Cloud
    • Web Platform

  2. AI Lead · Ecossistema 300 Franchising

    Nov 2025 – Apr 2026

    • Led a team of 4 engineers building internal AI for 170+ employees.
    • Architected multi-agent systems and RAG pipelines.
    • Operated 2 on-premise LLMs (32 GB of VRAM).
    • LLMs
    • RAG
    • Multi-agent Systems
    • On-premise Inference

  3. CTO · Lumus Tech

    Aug 2024 – May 2026

    • Led the build of a SaaS identity-verification platform.
    • 9 deep learning models developed, 3 in production.
    • Deepfake detector with 90% accuracy.
    • Curated a proprietary dataset of ~200 GB / 1M images.
    • Deep Learning
    • Computer Vision
    • Cloud
    • MLOps

Research & Academia

Master's

Computer Science

Mar 2025 – present

State University of Maringá (UEM)

Advisor: Prof. Dr. Marcos Aurélio Domingues.

research project

Hybrid Visual Feature Extraction for Deep Learning-Based Fashion Product Recommendation

Convolutional Neural Networks (CNNs) and Vision Transformers (ViTs) are used as visual extractors in fashion recommendation, but always in isolation. This project investigates whether hybridizing them, through late fusion of embeddings or hybrid architectures that combine convolution and self-attention, produces more informative visual representations and, consequently, more accurate recommendations than each family used alone.

View Prism VRec on Zenodo →

Engineer

Production Engineering, Software Emphasis

2019 – 2024

State University of Maringá (UEM)

academic projects

  1. Applying Data Mining Techniques to Acquire Contextual Information for Point-of-Interest Recommendation

    Mar 2023 – Feb 2024

    Mined point-of-interest data to extract contextual embeddings via neural networks, then compared recommendation techniques for point-of-interest systems.

    Advisor: Prof. Dr. Marcos Aurélio Domingues. Co-advisor: Profa. Dra. Gislaine Camila Lapasini Leal.

  2. Changing Electricity Consumption Habits with Gamification and Recommendation in an Educational Game

    Sep 2022 – Sep 2023

    Designed a mobile educational game combining gamification with recommender systems to raise awareness and change behavior around efficient electricity use.

    Advisor: Prof. Dr. Marcos Aurélio Domingues.

  3. A Mobile App for Exploring Artist Relationships in Music Networks

    Sep 2021 – Aug 2022

    Built a mobile app that lets users explore artist relationships as a graph, stream music for free, and get recommendations based on their interests.

    Advisor: Prof. Dr. Marcos Aurélio Domingues.

  4. A Web Platform for Visualizing and Interacting with Music Networks

    Sep 2020 – Aug 2021

    Built a web platform to visualize and interact with music networks as a graph, with free playback and personalized recommendations.

    Advisor: Prof. Dr. Marcos Aurélio Domingues.

research interests

  • Efficient Vision
  • Domain Adaptation
  • Self-Supervised Learning
  • Multimodal Vision (RGB + LiDAR/IR)
  • Industrial Anomaly Detection
  • Efficient VLM Fine-tuning (LoRA)

Projects

Cattle Counting from Drone Imagery

computer vision for agro · Kasco AI

Counting herds by hand from aerial footage is slow and error-prone. I work on instance segmentation for automatic cattle counting in drone videos, with a messaging pipeline for processing at scale.

  • Deep Learning
  • Instance Segmentation
  • Detection Algorithms
  • Message Queues

Field Operations Platform

web platform · Kasco AI

Counting operations involve pilots, flights and clients spread across farms. I work on the web platform that orchestrates field operations end to end, from flight planning to the final report.

  • Full-stack Web
  • Cloud Infrastructure
  • Async Processing

Identity Verification Platform (Lumus)

SaaS · Lumus Tech

Digital onboarding needs to catch fraud without blocking real users. I led the build of an identity-verification SaaS with deepfake detection (90% accuracy) and 3 models in production.

  • Deep Learning
  • Computer Vision
  • Cloud
  • MLOps

Publications

Contact

Open to consulting, engineering roles and research collaboration.

contato@lucas-couto.com
LinkedIn GitHub Download CV

Brazil · working remotely worldwide