## Introduction **LineItem** ([lineitem.co](https://lineitem.co)) is an AI-native commercial property management platform. It lets AI handle the operational busywork of managing commercial properties — finding and vetting vendors, running RFPs, generating contracts, checking compliance — so manager reps can simply review and approve. I designed and built the product end to end: interface design, architecture, backend, data extraction pipeline, and search infrastructure. Every line of code was written with Claude Code, not through loose "vibecoding" but with tightly controlled, AI-assisted development. The two biggest challenges on this project were: 1. Efficiently extracting complete and precise vendor data from more than 3M web pages on a tight budget. 2. Designing an intuitive way to integrate an agentic assistant into the UI. ### Design principle: review and approve One of the guiding design principles is **review and approve**. While roughly 95% of the operational work is automated by AI, the manager reps still review and approve everything. That's partly down to the current maturity of the technology — and partly because this industry carries a healthy distrust of automation. Keeping a human at the final gate makes the automation trustworthy enough to actually adopt. ### Agent architecture Under the hood, LineItem runs on close to 60 agents — around 20 high-level agents coordinating roughly 40 specialized sub-agents. Combining that structure with strictly structured input and structured output gives a very high level of predictability at a very low computational cost. It's what made it feasible to visit and extract vendor data from almost 3M web pages affordably, and to generate contract documents with a near-deterministic outcome. Note: This project was built in 2025. Some of the technologies and methodologies used are already obsolete. --- ## Product Tour **From line item to purchase order — automated, tracked, signed, and handled.** Each feature is an agent, or a chain of agents, doing the operational busywork, with the manager rep reviewing and approving the result. **Proposals** — generate RFPs with AI in minutes, share them with colleagues, and publish.

3D Modelling

Renovation

Capital One Labs