
Khorshid
Full-stack commerce platform with AI-assisted product onboarding.
- Node.js
- Express
- MongoDB
- MySQL
- LLM
- Problem
- Operators were drowning in manual order entry, ad-hoc invoicing in spreadsheets, and a product catalog scraped from dozens of suppliers with inconsistent categories.
- Solution
- An Express + MongoDB platform unifying orders, transactional invoicing on MySQL, and an LLM-powered autofill that maps messy scraped categories to the internal taxonomy.
- Technical challenges
- Designing a transactional invoicing flow across two databases without losing ACID guarantees on the financial path.
- Keeping the LLM mapping cost predictable while still recovering from low-confidence predictions safely.
- Building a Zod-validated boundary so dirty scraped data never reached the order pipeline.
- Outcome
- Cut catalog onboarding time from hours to minutes and replaced spreadsheet invoicing with a single, auditable workflow.
