Hacking Together an AI-Powered Resale & Recycling Workflow

Aside from my developer dayjob, I’m co-owner of a junk removal and recycling company. We pick up unwanted items, sorts them for disposal, donation, or recycling.

For the last few months, we’ve been throwing a mix of generative AI, computer vision, and cloud tech at a pretty gnarly problem—how to streamline everything from customer intake to reselling upcycled goods. Here’s what we’ve cooked up:

•	Photo-Based Job Evaluation: We use Hugging Face inference endpoints running on GPUs to clean up and segment images, helping us estimate object volumes for collection jobs.  
•	LLM-Powered Product Descriptions & Inventory Sync: GCP endpoints manage our LLM calls, turning item photos into marketplace-ready product descriptions and syncing them with Shopify.  
•	Live Transcription & Sales Insights: We’re experimenting with LiveKit for real-time call transcripts, helping phone operators with sales insights and CRM automation.  
•	In-House Call Transcription System: We built our own system to transcribe and structure customer calls for CRM purposes—no reliance on third-party services.  

The demo will dive into the nuts and bolts—how we’ve wired up Firebase, GCP, OpenAI, Anthropic, Hugging Face hosting, Twillio and LiveKit to make all this work. I’ll touch also on how using Cursor.ai really sped up our development.