Momentum

Momentum aimed to eliminate repetitive software processes, without ripping out existing systems and processes. We believe that far too many people spend their valuable time going through tedious digital workflows, leading to burnout and less time spent on meaningful work. We started with real estate brokerages, who we found had many mundane processes surrounding real estate transactions, of which they see hundreds or even thousands per week.

To solve this problem, we built a generalized platform that lets a user specify natural language instructions for navigating websites as well as downloading and processing documents. These instructions were fed to AI agents that could then perform complex, multi-step tasks on any cadence, at scale.

On the technical side, I built the entire software system with my co-founder. Some of the highlights:

• An RPA bot, which used Selenium on a cloud server in our cluster to execute actions provided to it.
• An LLM-based “planner”, which produced explicit actions for our RPA bot to execute based on several inputs such as a labeled screenshot of the webpage, the DOM, etc.
• A “perception” system, which would clean raw inputs provided by the RPA system and prepare them for use by the planner
• A “block” orchestration system that let us run self-contained instructions (blocks) in series, each with a variety of potential functionalities, and could use prior block results as inputs
• Several types of blocks that did things like download and analyze documents, run an RPA bot cycle, or compare a prior result to a user input