CompanyCam

CompanyCam is the photo app contractors use on job sites: 140,000+ contractors take photos, write reports and send them to customers. I was on the Workflows & Outputs team from November 2023 to August 2025, working in a large Rails monolith.

Pages AI

Pages is CompanyCam's document builder. Contractors use it to turn job photos into reports and other documents for their customers. Writing those on a phone at a job site is slow.

I led the Pages AI assistant from the first idea to production. A contractor types or talks into their phone on the job site and gets finished documentation back. It grew to about 5,000 requests a day. With the assistant, the share of active companies using Pages went from 4% to 11%, and customers told us it saved them 2 to 5 hours a week.

We measured adoption by company, not by user. One company can have a lot of users who never make a document, so the per-company number was the more honest one.

The shared AI layer

Every AI feature needs the same plumbing: picking a model, prompting it, calling tools, handling failures.

So I built shared Rails abstractions on top of RubyLLM, and they were adopted across the product. RubyLLM doesn't care which provider you use, and we shipped on Anthropic, OpenAI and Groq. Voice commands, context-aware invoice generation and tool-calling workflows all ran on that layer.

I also worked on evals and autonomy gating for the AI features: checking output quality and deciding when the system could act on its own.

Search with RAG

I worked on the RAG pipelines too. We indexed the backend records into pgvector so search could understand natural language, not just keywords. That included photos: type "cat" and you get the photos of cats.

Other work

• Share Link (Rails, React/TypeScript): any asset a contractor has, they can share securely with clients and payers who aren't on CompanyCam.
• PDF exports: cut file size 10x with streaming, compression and async processing on Sidekiq and S3, for 10,000+ exports a day.