Juan – AWS, React, Typescript

A conversational AI companion designed to support users’ mental well-being through guided, empathetic interactions. The system uses a custom-built orchestration layer developed in Node.js, enabling fine-grained control over conversation flow, context management, and dialogue logic. This architecture allows dynamic routing of user inputs, adaptive responses based on emotional state, and flexible expansion of conversational scenarios while maintaining safety and consistency across interactions.

• Built a custom LLM orchestration layer prior to the availability of tools such as LangSmith, enabling full control over conversation flow and system behavior;
• Led and coordinated the development team, overseeing architecture decisions and delivery milestones;
• Created internal tools that allowed product stakeholders to debug conversations, validate changes, and safely test updates;
• Designed collaborative workflows that enabled close cooperation between engineering and product teams on prompt development and iteration;
• Developed and maintained APIs supporting core platform functionality and internal integrations.

A web application designed for large labeling teams to efficiently navigate video and image datasets and annotate each frame. The platform streamlined the labeling workflow, ensuring consistency and accuracy across contributors, and produced high-quality datasets for training computer vision models.

• Developed and maintained the web application, ensuring stability, performance, and usability for large labeling teams;
• Performed a core refactor to migrate the codebase to fully React-based components with TypeScript, improving maintainability and scalability;
• Implemented interactive annotation tools, including paintbrush and eraser functionality, allowing precise frame-by-frame labeling of images and videos;
• Optimized user interactions for efficiency, including smooth rendering of large datasets and responsive UI for high-volume labeling workflows;
• Collaborated with designers and product managers to refine features and enhance the overall labeling experience.

Worked across multiple teams on a web platform for running A/B tests on features leveraging user-generated content data. Developed and maintained internal tooling to support experiment setup, monitoring, and analysis, enabling data-driven decisions and optimization of product features.

• Increased website sales by designing and implementing A/B tests to optimize key features and user flows;
• Enabled service agents to improve performance metrics by providing actionable data insights and streamlining BI report preparation;
• Reduced time spent on report generation and data analysis, improving operational efficiency and supporting data-driven decision-making

A wizard application enabling sales representatives to visit clients and design aircraft in real time, without requiring an internet connection. The app handled 4+ GB of assets to generate multiple interior and exterior views, while maintaining smooth, high-performance rendering. Built with PhoneGap and vanilla JavaScript for maximum efficiency, the application also leveraged accelerometer data to provide an augmented reality view of the aircraft for an interactive client experience.

• Led the development team, coordinating tasks, architecture decisions, and delivery milestones;
• Built multiple prototypes to evaluate and identify the most efficient rendering strategies;
• Designed and executed experiments to optimize performance on limited hardware resources, including iPad 2;
• Developed the core application using vanilla JavaScript, ensuring high performance and responsiveness.