DJ Ops — Paperwork Digitizer

One-liner: A mobile-first PWA that photographs a physical event contract or run-of-show sheet and extracts the details into a searchable database with Claude Vision — so “where’s that paperwork?” stops being a question I ask mid-event.

Origin (the Build Loop)

Because I DJ 25–40 weddings and events a year, I kept hitting the moment on the floor where I needed a detail off a signed contract or run-of-show and it was a photo buried in my phone or a sheet in a folder in the car, so I built a PWA that snaps the paperwork, reads it with Claude Vision, and files the extracted fields into a searchable database, that puts every event’s details one search away, live, from my phone. It taught me that the photograph-it-and-extract pattern is reusable — I later built the same core into thrift-lens.

The problem

Working DJs run on paper: contracts, run-of-show sheets, pay details, timelines. In the moment — mic in one hand, next song cued — you don’t want to be scrolling a camera roll or digging through a folder for the one number you need. I wanted to photograph the paperwork once and have it become structured, searchable data I could pull up instantly at the next event.

What it does

• Scan paperwork — photograph a contract or run-of-show sheet from the phone.
• AI extraction — Claude Vision reads the image and pulls the structured details (event info, timeline, pay) into the database.
• Event management — manage events and track pay across them.
• Searchable history — every digitized document is queryable, so any detail is one search away on-site.
• Installable PWA — works like a native app on a phone, built for use in the field.

How it’s built

• Stack: React + TypeScript + Vite (frontend, port 5000), Express + tsx (backend, port 3000), Drizzle ORM + PostgreSQL, @anthropic-ai/sdk for Claude Vision; heic-convert (handles iPhone HEIC photos); p-limit / p-retry for resilient batch processing; deployed on Replit with its identity system for auth.
• Notable engineering:
  • HEIC conversion — iPhones shoot HEIC; the pipeline converts before sending to Vision, a real-world detail most demos skip.
  • Resilient extraction — p-limit + p-retry handle rate limits and transient failures on batch scans rather than failing the whole upload.
  • Vision-to-schema — the hard part is mapping messy real-world paperwork to a clean Drizzle schema reliably.
• Architecture: phone camera → HEIC convert → Claude Vision extraction (rate-limited/retried) → Drizzle/Postgres → searchable PWA.

Proof points

• Purpose-built for a real workflow Alex runs 25–40 times a year.
• Handles the actual field conditions (iPhone HEIC, flaky venue connectivity, batch scans).
• Same Vision-extraction core later reused in a completely different domain (resale) — proof of a repeatable skill.

What to show

• Demo: Deploy to a public URL (repo is public). Seed with a sample (non-real) contract so a visitor can scan → see extracted fields → search. Use a dummy document, not a client’s real contract.
• Visuals needed: the scan/capture screen; the extracted-fields result; the search view. Blur or fake any real client/PII.

Cross-links

• Direct sibling of thrift-lens — same photograph-it-and-extract Claude Vision pattern, different domain. This pair is the portfolio’s clearest “reusable pattern, not one-off” evidence.
• Complements the off-repo Node.js/docx run-of-show generator and the wedding timeline planners in the weddings cluster.

Case-study angle

The tool that most literally captures the thesis: a working DJ got tired of hunting for paperwork on the floor, so he built a Vision-AI app to make it searchable from his phone — then reused the exact same engine to price thrift finds. One pattern, two businesses.