Henrique Alcantara Leite

A centralized, predictive command system designed to unify Canada's historically fragmented wildfire intelligence. By integrating data from provincial agencies, federal systems, and satellite networks into a single real-time platform, FireGrid enables fire managers to make data-backed, life-or-death deployment decisions.

Engineered a high-concurrency tactical command center using FastAPI and AWS DynamoDB to ingest and fuse live fire data from NASA FIRMS satellite hotspots with CWFIS/NRCan registries and Open-Meteo weather streams. Developed a predictive burn probability engine using XGBoost trained on CFFDRS indices. Implemented a Tactical RL Agent to autonomously score geographic choke points and recommend optimal interagency asset deployment.

🏆 2nd Place Overall at CXC 2026 AI Hackathon. Best Use of Backboard.io.

A manipulation-aware analytics platform for prediction markets that identifies "smart money" signals by profiling individual trader accuracy. Synthesizes a SmartCrowd Probability by weighting beliefs based on historical calibration, persistence, and domain specialization.

Engineered a real-time ingestion pipeline to process Polymarket trade sequences, utilizing a hierarchical Bayesian shrinkage approach to calculate per-wallet trust weights. Integrated advanced 3D WebGL visualizations using Three.js, including a PCA-based double helix to detect market regime shifts and a 3D lattice to map divergence magnitude in real-time.

An AI-powered conversational shopping assistant designed to bridge the gap between product discovery and the final checkout. By leveraging the Universal Commerce Protocol (UCP), it creates a "Jarvis-like" experience where users move from a vague natural language intent to a fully realized Shopify cart in a single interaction.

Engineered a sophisticated agentic architecture using LangGraph and FastAPI to orchestrate stateful decision-making loops, utilizing the Model Context Protocol (MCP) to standardize tool-calling and secure communication with the Shopify Storefront API. Implemented a personalization engine using MongoDB to store user-specific search history and preferences, enabling the agent to autonomously navigate the checkout flow.

RBC Design Thinking Project (Western University).

A prevention-first wildfire defense concept combining an exterior sprinkler barrier, real-time environmental monitoring, and computer vision to detect smoke and embers early and trigger protection automatically.

Built a Raspberry Pi sprinkler MVP with sensor logging and relay control, plus a YOLOv8-based fire and smoke detection pipeline exposed through a FastAPI server for live status and video streaming. Designed a homeowner dashboard UI in React (Vite) to surface alerts, manual overrides, and localized risk data for insurers.

🏆 2nd Place AWS Track — Best Use of DynamoDB at Hack The North.

A real-time, gamified stock market simulation designed to teach financial literacy through interactive gameplay. The platform features a live order book and simulates market volatility using AI-generated news events.

Developed a serverless, event-driven architecture using Amazon DynamoDB Streams to handle real-time state changes and order matching with low latency. Integrated the Cerebras Inference API to generate dynamic, hallucinated financial news headlines that instantly impact stock prices, creating a realistic trading environment.

A user-friendly web app designed to help users assess the credibility of websites in an age where AI-driven scams and deceptive ads are on the rise. By leveraging a Human-In-the-Loop agent workflow, AdLumen analyzes a site's images, domain, and content to determine if it may be fraudulent or misleading. Users simply input a URL, and the app provides an easy-to-understand assessment of the site's trustworthiness.

The frontend is built with Next.js and TailwindCSS, while the backend uses FastAPI and Python to orchestrate the agent and process data. MongoDB is used for storing scan results, with BSON for data interchange. The project also utilizes Gemini for AI capabilities and is written in TypeScript and Python.

🏆 1st Place at Toronto Tech Expo — Best Real-World Use Case, sponsored by Slalom.

AI-powered system designed to detect AI-generated voice clones used in phone scams and cyberattacks, leveraging machine learning techniques to analyze audio files and identify synthetic speech with high accuracy.

Led a team of 5 developers to build a full-stack cybersecurity solution that detects AI-generated voice scams in real-time. Processing raw audio with Mel spectrograms & LFCC features, training a ML model to classify real vs. cloned voices.

Using random forest, the model achieved over 87% accuracy. Won the Best Real-World Application Award at Toronto Tech Expo, sponsored by Slalom.

Developed as part of the Automotive Innovation Challenge hosted by Western University Engineering Faculty and sponsored by General Motors.

Real-time road user detection system using thermal imaging technology, specifically designed to improve pedestrian and cyclist safety in low-visibility conditions by detecting and classifying pedestrians, bicycles, and vehicles.

Implemented Faster R-CNN with ResNet50-FPN backbone, achieving 87% mean average precision and high detection accuracy across different classes.

A machine learning-based system for predicting bus arrival times in London, Ontario. This project uses a Bidirectional LSTM neural network to provide accurate predictions for various bus routes.

Led a team of six developers in collaboration with the Mobility Technology (MoTech) research group under Dr. Yili Tang at Western University.

Integrated real-time bus location data from the City of London transit portal and incorporated weather APIs to refine prediction accuracy.

A virtual pet simulation game developed using Java 23+, featuring interactive pet care mechanics and parental controls.

Built with Java Swing and AWT for the user interface, the project implements data persistence through CSV files and includes comprehensive unit testing with JUnit 5.

The game demonstrates our team's ability to create engaging applications while maintaining clean code architecture and following agile development principles.

Modern, responsive personal portfolio website built with Next.js and featuring advanced animations and interactive elements.

Showcases professional experience, projects, and technical skills with a dark theme design inspired by modern web applications.

Implements smooth scroll animations, flickering grid backgrounds, and shimmer button effects for an engaging user experience.

Developed a Python-based tool for options pricing, profit and loss (PnL) calculation, and real-time financial data integration.

This project uses the Black-Scholes model to price European call and put options, along with custom-built modules to track and analyze option portfolios.

By integrating with the Yahoo Finance API, the tool ensures accurate and real-time market data to enhance the precision of options pricing and PnL calculations.