Matthew Don!

Built a localized platform for AI model access, cost-aware routing, API credits, and media generation. The product combines a server-rendered Next.js application with real-time Convex data and workflows, WorkOS organization identity, and a model-routing layer I built to manage API tokens, token usage, and credits. It also connects payment events, S3/R2 media storage, customer documentation, and Twenty CRM/editorial operations into one production system.

Built a Telnyx + WebRTC softphone and power‑dialer. I created isolated Docker services exposed through ngrok to handle Google Maps/Google scraping, Instagram messaging, and Discord community outreach for mass booking workflows. I added an SMS follow‑up module using Convex Agent components to keep booking follow‑ups organized with a human‑in‑the‑loop rather than fully automated replacements. The platform includes credits, multi‑workspace RBAC for agencies, real‑time activity views, call recordings, and AI transcription analysis. It began as a personal tool and reached 35 users in one month.

While working on Diala, I met Hu Yu Dong in Vietnam when he was exploring building a factory. He invited me to China to work directly for Jianxi Plastic Factory and to build a similar scraping and messaging system for his China-based sales team and himself. There, I learned how regional, provincial, and international trade is handled in China and built bespoke systems to automate the monotonous outreach tasks involved in running a factory. That work sparked my interest in sourcing, so I created a sourcing company using the same stack and expanded it with additional channels (LinkedIn, Reddit, SMS, Instagram, and Facebook) via my outbound tool Pivora. This helped me book meetings with coffee shop and boutique restaurant owners in California to replace their current suppliers with factories I partnered with in China.

While building my sourcing company, friends worldwide asked me to buy items outside my core offerings (PVC, pumps, and plastic utensils). Requests ranged from clothes and toys to tools, so I built Gambuy in Vite with an ingestion engine that pulls listings from 1688, Taobao, Weidian, and competitor agents like Gtbuy, Mulebuy, and Kakaobuy. I aimed for a similar offering but added gamified incentives to drive return visits and give users chances to win shipping discounts. I also reused the clothing-replacement tooling I developed with Jane at Fitexto to build a pipeline for AI-assisted virtual try-ons: a vision model analyzes the garment and the user's uploaded photo so customers can preview fit and style before purchasing.

Approached by friends Martinus and Anders to build a site that places Oil & Gas engineers into roles across Norway’s energy sector. We used Jina’s segmenter and reranker to parse CVs and route strong candidates into an interview pipeline, combining Martinus’ Meta Ads acquisition, my embedding/model work, and Anders’ 8 years of upper-management industry insight. The system shortlists candidates and forwards them to management for review; successful hires generate commission. I built the full software stack as the sole developer and onboarded hiring managers on how to use it.