Loading...
Cantaloupe AI is a comprehensive next-generation recruitment and job application platform that leverages AI and advanced automation to streamline the entire hiring process. The platform combines AI-powered candidate screening using resume parsing and voice interviews (powered by Retell and Vapi), integrated multi-channel communication via Twilio, SendGrid, and Resend, and bi-directional synchronization with popular ATS/HRIS platforms through the Unified API and Merge API. The architecture features multiple implementations: a high-performance Go backend service with SvelteKit frontend (Ahuramazda/Mithras), and a comprehensive Next.js full-stack application (Union) serving as the main production platform at trycantaloupe.com. The system reduces hiring time, improves candidate quality, and delivers measurable ROI with implementations demonstrating $4,000+ in monthly savings for mid-size operations through intelligent automation and data-driven hiring insights.
Cantaloupe had started as a mobile-first platform in May 2024. I joined on August 7, 2024 and immediately started contributing to the React Native/Expo mobile apps for job seekers and businesses. Worked with serverless AWS Lambda backend. Built UI components like PersonalityQuestions and sliders.
Two weeks after joining, started Svelte experiments (August 20). First tried SvelteKit in mongoose repo for custom development, then experimented with Bubble.io no-code platform for faster iteration, but returned to Svelte for more control. Added Go backend to replace serverless during this period.
Matthew started union repo on June 29, 2025. Teamed up with Josh and Matthew to build the production web platform using Next.js. Migrated learnings from React Native and Svelte experiments. Implemented Voice AI interviews (Retell), calendar integration (Nylas), ROI analytics, and comprehensive business dashboards.
Continuously improving the platform with enterprise features, multi-language support (English/Spanish), Twilio messaging, automated workflows, and performance optimizations. Production deployment on Vercel serving real businesses and job seekers.
A production-ready computer vision model that detects and classifies hands, arms, and non-hand objects in real-time with 96% accuracy.
Local-first memory layer for AI coding agents. 14 MCP tools, 268K+ indexed chunks, hybrid semantic+keyword search, local LLM enrichment. pip install brainlayer.
Local voice input/output layer for Claude Code. 7 MCP tools, 5 voice modes, whisper.cpp STT (~300ms), edge-tts, session booking. bunx voicelayer-mcp.
A comprehensive e-commerce platform for handmade candles featuring a React-based storefront with shopping cart, favorites, filters, and a Node.js backend with MongoDB. Fully responsive design with admin dashboard for order management.