Pedro – Typescript, React, Python

A production-grade Generative AI platform built on Retrieval-Augmented Generation (RAG) and LLM tool calling (function calling) to leverage enterprise data. The system equips LLMs with specialized tools to interact with internal APIs and databases, providing employees with context-aware, grounded responses from vast knowledge bases. By enabling models to validate information in real time, the platform effectively eliminates hallucinations and streamlines complex information discovery across the organization.

• Designed and developed serverless AI microservices using Knative on Kubernetes, implementing scale-to-zero policies to optimize resource consumption and ensure high availability for fluctuating GenAI workloads.
• Designed and developed end-to-end RAG pipelines using Python and TypeScript, implementing advanced chunking strategies and embedding models to integrate retrieved context with LLMs.
• Designed and developed a dual-layer backend API architecture, orchestrating public-facing endpoints for the UI alongside internal microservices dedicated to document chunking and embedding generation.
• Developed functional proofs-of-concept for emerging features, translating high-level business requirements into scalable technical prototypes for future production phases.
• Acted as the technical focal point, bridging the gap between architectural design and implementation while mentoring developers on scalable backend patterns and robust system integration.
• Led critical-path troubleshooting, intervening in frontend performance issues and complex architectural defects to ensure the platform met enterprise stability standards.
• Acted as a DevOps bridge by designing and developing CI/CD pipelines using GitHub Actions, Bash, and Terraform (IaC) to automate the deployment of microservices and infrastructure across multiple environments when dedicated DevOps support was unavailable.

A multimodal agentic platform that automates corporate assessments by orchestrating AI agents to analyze complex documentation and generate strategic reports based on business-defined mandates.

• Developed a graph-based orchestration engine using LangGraph, implementing state management and cyclical reasoning to guide agents through multi-stage assessment workflows.
• Engineered a context-aware reasoning layer using prompt and context engineering, ensuring AI agents strictly adhered to predefined business rubrics through specialized node-level instructions.
• Designed and implemented robust AI guardrails and Responsible AI practices by building validation layers that prevented hallucinations and ensured all generated assessments were grounded in provided evidence.
• Implemented multi-modal ingestion microservices to process unstructured text and audio data, enabling the platform to cross-reference interview transcripts with written corporate policies.
• Designed an interactive post-assessment user chat interface, allowing users to query the LLM across both raw inputs and generated assessment reports for deeper insights.
• Directed the technical evaluation of agentic frameworks, spearheading LlamaIndex proofs-of-concept for RAG and autonomous agents to define a roadmap for future capability expansion.

An AI-driven search platform for KPMG Canada that enables semantic document discovery using vector embeddings. The system processes unstructured data through a scalable ETL pipeline, allowing users to find information based on context and meaning rather than simple keyword matching.

• Architected and implemented a scalable, asynchronous ETL infrastructure using Knative on Kubernetes, leveraging Azure Storage Queues for resilient, message-driven document processing.
• Optimized cloud performance and scaling by conducting stress tests with Artillery and using the results to fine-tune Knative configurations, significantly reducing cold starts and improving auto-scaling behavior.
• Collaborated closely with the DevOps team to ensure serverless environments met both infrastructure stability requirements and application-level scaling expectations, delivering a highly available system.
• Designed and developed high-scale ETL pipelines in Node.js, TypeScript, and Python to ingest and transform unstructured data into vector embeddings for Azure Cognitive Search.
• Designed and developed a dual-layer API architecture, delivering public-facing REST APIs for the React frontend alongside internal microservices (e.g., OCR services) to support complex data extraction workflows.
• Served as the technical gatekeeper for the platform’s RESTful APIs and ETL services, enforcing strict data validation and reliability across the serverless environment.
• Developed an intuitive React frontend to help users visualize and interpret complex semantic search results and supply chain impact data.
• Resolved frontend rendering bottlenecks by implementing virtualization techniques, enabling the UI to display thousands of data points with high performance and no perceptible lag.
• Ensured seamless system integration and data integrity by implementing a comprehensive testing suite using Jest (unit) and Playwright (E2E), covering the full lifecycle from data ingestion to the UI.

A high-performance B2B2C neurofeedback platform that enables remote brainwave activity (EEG) in real-time through wearable neuro-devices.

• Architected and implemented a scalable, serverless cloud infrastructure using AWS Serverless services to support high-frequency bio-data ingestion from neuro-wearables with sub-second latency.
• Designed an event-driven communication layer using Node.js, TypeScript, and RxJS, enabling real-time processing and broadcasting of brainwave data to clinician dashboards.
• Engineered high-performance data visualization components in React to render real-time EEG brainwave streams for clinical analysis.
• Designed and delivered end-to-end features for patient management, session history, and progress monitoring, covering both RESTful APIs and the React frontend.
• Integrated Stripe for payment management and ensured high code quality through comprehensive unit testing (Jest) and end-to-end testing (Playwright).
• Provided cross-team technical support by contributing to the React Native mobile codebase, resolving persistent issues, and mentoring the mobile team on JavaScript best practices to keep the project on schedule.
• Collaborated closely with hardware SDK developers to troubleshoot and patch Python code, resolving critical data transmission issues.

An open source web server and client application for aggregating content from blogs about open source technologies.

• Migrated the Frontend from Gatsby JS (Javascript) to Next JS (Typescript) • Designed and implemented a new UI/UX • Implemented unit and end-to-end tests. (Jest, Playwright)

An accessibility-focused Visual Studio Code theme designed to enhance the development experience for engineers with color vision deficiencies (protanopia, deuteranopia, and tritanopia). The project bridges the gap between the aesthetic of the official default theme and the functional necessity of distinguishable color palettes for syntax highlighting. The extension has reached over 8,000 installs with a 4.5-star rating.

• Developed a custom color token mapping system using the VS Code Extension API to deliver high-contrast syntax highlighting optimized for colorblind accessibility.
• Analyzed WCAG (Web Content Accessibility Guidelines) and color theory to select color palettes that preserved code readability while maintaining the original client`s visual identity.
• Optimized the theme’s JSON structure to ensure zero performance overhead and seamless integration with existing VS Code UI components.
• Managed the end-to-end publishing process for the Visual Studio Marketplace, including documentation (README), versioning, and extension branding.