Lucas – Python, AWS, Typescript

The platform provides automated alerts and analytics data, monitoring cameras to generate alerts, statistics, and insights tailored to the client's needs, enabling timely action. For instance, on a highway, if an animal is detected on the road, the system generates an alert with video evidence and sends it to the web app, allowing personnel to promptly address the situation and remove the animal from the roadway.

• Led the implementation and maintenance of OMNICOG's computer vision-oriented code using Python, C++, Apache Kafka, Ignite, Faust, Deepstream, and Triton inference servers;
• 20% of my time in this project is spent helping the web app team with the new features. The stack is typescript/mySQL/React with Kubernetes deployment;
• Architected the project's new structure focusing on modularity, improving performance by 37%;
• Reduced stream processing latency by 73% by optimizing the use of Apache Kafka and Faust;
• Improved model inference accuracy by 16% by using Deepstream and Triton inference servers;
• Secured a contract with the Brazilian government to deploy the computer vision system in 4 states, with plans to expand to more states in the future;
• Implemented GPU-accelerated algorithms for real-time video processing using NVIDIA DeepStream;
• Enhanced the performance of inference models on GPUs, achieving significant latency reduction;
• Utilized OpenCL for specific parallel processing tasks within the project;
• Leveraged NVIDIA CUDA and LLVM for optimizing computationally intensive tasks;
• Utilized AWS EKS to run our containerized inference services;
• Created a secured inter-service communication via custom AWS VPC setups with VPNs.

The product is a VR platform compatible with both Android and macOS glasses, allowing users to input a wide variety of content, including apps, videos, and games. All usage data is collected and stored in a database, providing administrators with full access to various metrics. Additionally, the platform supports both online and offline content delivery and features a built-in AI helper bot to assist customers in navigating the solution.

• Oversaw and managed a team of 3 web developers, responsible for the development and maintenance of all web applications on the Azure and Heroku platforms;
• Developed and maintained APIs using JavaScript and Node.js, responsible for handling over 75k requests per day;
• Created new features for the CMS and LMS using Python and Django, increasing user engagement;
• Found and fixed bugs from legacy code, reducing the number of production incidents by 70%;
• Wrote and reviewed maintainable Node.js code, following best practices and coding standards;
• Configured and maintained the Azure and Heroku environment, ensuring optimal performance and scalability;
• Wrote new tables, scripts, and queries for PostgreSQL, optimizing database performance by 30%;
• Served as a DevOps engineer when required, responsible for database maintenance, creating and managing an OpenShift cluster, and maintaining legacy Kubernetes environments;
• Configured AWS VPC and VPNs settings to secure our cloud connectivity to be able to use AWS EKS for container orchestration and SQS for messaging;
• Integrated AWS SQS to run our async messaging between our web app components, so we could have decoupled communication for features like the CMS and LMS.

This organization specializes in artificial intelligence, data science, and cybersecurity and is dedicated to developing end-to-end solutions that drive digital transformation and innovation for businesses. Key projects involved enhancing user experience and operational efficiency within Samsung’s Windows ecosystem through deep software-hardware integration, including adapting mobile SIM card modems for notebooks, improving the Samsung Gallery app for faster performance, and addressing automotive industry challenges by replacing faulty hardware to eliminate frequent recalls.

• Developed and maintained a Windows kernel driver using C++, increasing system performance by 25%;
• Implemented smart pointers and containers in C++ to overcome the lack of STL support in kernel mode;
• Worked with telecommunication/network protocols on a daily basis, developing and optimizing code for high performance and reliability;
• Used Git, Bitbucket, GitHub, and Perforce for version control;
• Automated tasks using Python and Windows shell scripts, saving over 10 hours of manual work per week;
• Set up AWS VPCs to maintain secure and isolated communications across our systems.
• Integrated AWS SQS into our automated PDF processing workflow to queue events for real-time bug report handling.
• Engineered an advanced automated system for real-time processing of Samsung bug reports, significantly enhancing operational efficiency by automating data extraction and report generation;
• Trained and mentored newcomers on the kernel driver team, helping them to quickly become productive members of the team;
• Received a Performance Award for outstanding contributions to the team;
• Developed new features for the Samsung Gallery app for Windows, improving features and performance by 33%;
• Used C++ for image processing and Python for running and training the AI model;
• Improved one of the AI models with a new design using transformers, increasing accuracy by 17%;
• Developed and maintained software for an automotive embedded system using C and the specific assembler language for the electric coil controller;
• Automated 6 processes using Jenkins pipelines;
• Integrated NVIDIA CUDA for GPU-accelerated image processing tasks, enhancing performance;
• Developed and optimized GPU algorithms to support high-performance computing needs;
• Employed LLVM for optimizing kernel and user-space code;
• Implemented GPU-accelerated solutions using OpenCL for specific performance-critical tasks.

This organization is a UK-based charity focused on supporting individuals with communication challenges and complex needs. The project involved developing an engine utilizing machine learning to detect and classify eye movement into five different positions, using only webcams and smartphone cameras. This technology primarily serves the healthcare sector, aiding patients with severe movement limitations.

• Developed two distinct computer vision approaches for eye tracking: one utilizing deep learning techniques and another implemented using C++, Python, and OpenCV, in collaboration with a project manager and two back-end developers;
• Created a machine learning engine capable of detecting and classifying eye movements into five positions, leveraging webcams and smartphone cameras, achieving an impressive accuracy rate of 92%;
• Utilized PostgreSQL to efficiently store user data, enhancing the accuracy of the eye-tracking model.
• Committed all code to GitHub from the outset, ensuring the project remained open-source and accessible to the wider community.
• Collaborated with Google engineers to exchange knowledge and enhance the AI model's performance.
• After delivering the initial MVP, partnered with two engineers to develop a comprehensive web application using Node.js and React, making the solution open-source and freely available. Following delivery, the client implemented their own adjustments, resulting in a user-installable app.