GenAI Hackathon by Impetus & AWS
SmartSkin Emerging Leaders in AI Prize
Inaugural 2026 recipient of the SmartSkin Emerging Leaders in AI Prize, presented by SmartSkin Technologies and UNB Computer Science at the Atlantic AI Summit kickoff.
Projects
AI-powered video creation platform spanning a desktop editing studio and a web-based AI editor. The desktop stack combines a C++/N-API recorder (DXGI, Media Foundation, NVENC) with an Electron orchestrator, while the web app lets you describe what you want in plain English — powered by Google AI Pro, Veo 3.1, and Lyria — supporting chat-to-edit, voice control, auto captions, beat sync, and AI effects.
Engineering Notes
- Wrote a Win32-native recorder that swaps between NVENC, AMD VCE, and software fallbacks at runtime while sharing D3D11 textures with the Electron renderer through zero-copy ring buffers.
- Exposed OS hooks (desktop icon toggling, monitor/window selection, cursor/keyboard trackers) through a custom N-API bridge so overlays stay in sync at sub-frame latency.
- Spawned GPU-aware Chromium clusters for export; each renderer streams frames over Socket.IO to an FFmpeg bridge that stitches ProRes/H.264 outputs without blocking the main process.
- Chat-to-edit interface: type natural language instructions and watch the timeline update in real-time.
- Integrated Google Veo 3.1 for AI video generation and Lyria for music generation directly in the editing workflow.
- Voice control for hands-free editing with real-time feedback during preview playback.
Created a CLI + GitLab integration that measures the carbon footprint of every pipeline run. The wizard configures runners on AWS/GCP, wires electricity intensity data, appends CI jobs, and posts detailed GitLab MR comments with emissions budgets.
Hackathon Wins
Chrome extension that automates the web entirely offline using Gemini Nano and Chrome's built-in AI APIs — Prompt, Summarizer, Rewriter, and Writer. Voice-controlled, agentic browser automation with semantic memory and document RAG.
AI-powered ecommerce platform augmenting Square databases with chat, interactive charts, data insights, psychographic analysis, and voice-enabled chart explanations.
Serverless forest monitoring system using AWS Lambda, Sentinel-2, and SageMaker — tracks deforestation in real-time with NDVI indices and K-means clustering.
Turns noisy AWS logs into actionable incidents using TiDB vector search and agentic LLM. Ingest, embed, detect spikes, and route alerts—serverless, fast, safe.
AI-powered supply chain forecasting platform using LLaMA 3 models to turn raw ERP data into actionable, real-time business insights—cloud-native and production-ready.
A restricted local browser agent that uses Auth0 Token Vault to securely act across Google and Linear + more, with explicit permission boundaries and approval-aware delegated actions.
Built the official PHP/Laravel bindings for 1Password's Passage SDK, enabling seamless biometric authentication for PHP environments — filling a critical gap in the PHP ecosystem.
AI-powered database editor plugin that lets you run AI queries directly on table cells — fix grammar, translate content, condense text, and more using Meta's LLaMA model.
Work Experience
January 2025 - Present
- Slashed inference latency by 65% optimizing Elasticsearch queries and FastAPI endpoints for real-time threat detection
- Built agentic deep-report workflows for threat intelligence, improving report quality, editability, cost, and runtime performance
- Developed URL risk-detection models and sandboxed analysis pipelines using enriched web, DNS, screenshot, and page-content signals
- Cut CI/CD runtime by 75% through Docker layer caching and parallelized GitHub Actions workflows
- Hardened APIs to OWASP standards while driving test coverage from 40% to 85% across the stack
2022 - 2024
- Built real-time 3D video effects using TensorFlow.js and WebGL, shipped in official Twilio Video demos
- Authored technical tutorials and sample apps driving developer adoption of Twilio Video APIs
Blog
Shipping GitGreen: Measuring CI/CD Carbon in GitLab
What it took to build a CLI that instruments GitLab runners, posts MR carbon reports, and enforces emissions budgets.
GPU Performance Engineering: Lessons from Building Vidova AI
Deep dive into GPU optimization, frame-time analysis, and benchmarking methodology from building a video editing application.
Building Marionette: Can Small On-Device LLMs Handle Complex Tasks?
How I built a privacy-first AI browser agent that runs entirely offline, and what I learned about context engineering for small models.
Winning First Place at the 2023 Google and Square AI Hackathon
Sharing my experience and learnings from the hackathon, and the project I built.
Winning multiple hackathons in one year: 1Password, Outerbase (Cloudflare), and Google and Square
How I won multiple hackathons in one year, and what I learned from each of them.
Freelancing at Cyclic, Twilio, and DigitalOcean
A look back at my first year of freelancing and what I learned