Embedded Files
PROTOTYPE:
AI POWERED
CO-WORKER
INTRODUCTION
Built to reduce the friction of staying organized. The system creates a folder structure at the start of each project, puts files where they belong and has a a fully searchable database using plain text query. I went one step further and trained Claude on how to produce so that he could also assist with some of the more time-consuming parts of kicking off a project.
THE PROBLEM
Every production department runs on institutional memory — who we worked with, what we paid; what worked; what didn't. Right now that knowledge lives in folders nobody can find or in the heads of people who might leave tomorrow. I'm building toward a system where all of that data is queryable in plain English. Ask it anything: "show me every production we've done in Mexico and tell me who we worked with, how much it cost and show me the finished work."
THE BUILD
Built on top of Claude, the system uses Firebase to host the frontend and Cloud Functions to run the backend logic (API calls and Drive writes). The system is connected to Google Drive to host files. Claude has been trained to be a member of the production team and so he acts as more than a file retrieval service, Claude can also assist with breaking down scripts, putting together ballparks and building out pre-bid docs. He is a member of the production team.
RESOURCE MANAGEMENT ASSISTANT
INTRODUCTION
The agent was designed to do more than just recall data, it was build to be a member of the resource team, upload a project brief and Claude will check resource availability, ask follow up questions and make assignment suggestions based on a logic that includes career growth, performance and team fit, allowing the resource team to make faster matches and resource for teams they haven't worked with yet.
THE PROBLEM
With more than 300 resources spread across five resource managers coupled with the exit of key team members and new hires coming in, we needed a tool that would retain the knowledge from the previous team while also acting as a single brain across the entire pool of talent. The tool was built not a replacement for the resource team, but an additional member of the team that could apply it's own thinking and offer up suggestions.
THE BUILD
Built on top of Claude, the system was trained the same way we'd train any human member of the resource management team, first on human knowledge with the five resource managers talking to Claude about each person on their roster and answering questions about how they work. We then uploaded weekly reports for Claude to cross reference including capacity and utilization data from Workfront as well as Excel data about upcoming work
AP SCRIPT GENERATOR
INTRODUCTION
AP Scripts are required for network clearance of TV Commercials. A properly formatted AP script breaks down visuals on one column and audio in a second column. Previously a manual task that took up to 30 minutes per script, the AP Script Generator was able to generate a complete script in seconds.
THE PROBLEM
Building AP Scripts is a manual process of transcribing, shot by shot, all the video and audio elements into a Word doc. And it's required at the worst time in the production process, when a producer is trying to ship finished work by a deadline. Automating this task broke through a common bottle neck, saving time and also simplying the process so that anyone could create an AP Script, not just a producer., in a matter of seconds.
THE BUILD
Using a multi-agent tooo, pulling in video and audio transcriptions from Twelvelabs into Claude, who synthesizes both elements into a finished AP Script that is formatted correctly for network clearance.
ONBOARDING
AGENT
INTRODUCTION
Claude was onboarded as a new producer at Omnicom Production. I spoke to Claude for two hours, uploading supporting docs and answering his questions along the way. What resulted was an agent that could onboard new employees and be an always-on help desk for questions, eliminating hours of work for the resourcing team.
THE PROBLEM
It was a perfect storm for us a company, as new systems were being introduced, new employees were coming on board, and the team was understaffed to deal with it. The model was trained exactly like a new empoloyee, but was also told that after it was trained, it would do the onboarding.
THE BUILD
This was entirely built with Claude. The training resulted in a 5,000-word knowledge base explaining dozens of systems, outlining step by step instructions, as well as encapsulating the philosophy of working at the company.
Google Sites
Report abuse