About
Currently building rayloom.ai. I think AI safety is the most important problem humanity faces. I previously had longer timelines and planned on getting into the field in the future, but have since updated and think we are in crunch time over the next 2–3 years.
My main focus is how to get a joint US/China slowdown: roughly, AI 2040 plan A. I think the probability of a catastrophic AI outcome without a slowdown is too high for comfort. But I'm optimistic about automated alignment working given enough time.
In the past, I built AI video-editing and motion-graphics software, worked on AI voice agents, and built telephony and dialing products.
Background
-
Founder, rayloom.ai Sep 2025 - Present
- Solo-founded, currently just over 100k ARR.
- Built the entire product - suite of autonomous agents that handle the full content creation process including researching, scripting, video production and editing, and lead conversion
-
Co-founder and CTO, Picasso Nov 2024 - Aug 2025
- AI clipping and LLM video editor platform
- Built browser based LLM video editor and distributed processing/rendering infra using Modal, YOLO, and TalkNet
- Software produced 100,000+ edited clips
- Sole technical contributor, built the entire software
-
Product Team (Minority Co-Founder), AI Voice Agents May 2023 - Oct 2024
- AI voice agents for sales and customer service
- Designed large portions of the product UX and did prompt engineering to get voice agents to work with 4K context GPT-3.5 era models
- Worked directly with early customers in a forward-deployed role
-
Co-founder and Product Lead, Max Jul 2022 - Jul 2023
- Led a two-engineer team building a parallel and predictive dialer for inside-sales teams.
- Designed systems that optimized dialing using predicted pickup rates, answer probability, and time-of-day patterns.
Side projects
Autonomous Motion Graphics Editor Agent
An early autonomous video editor that does full motion graphics editing given an audio file. It uses Grok to chunk the script and plan the graphic for each scene, then creates the images, reflects on and improves them, writes the prompts for the animated graphic, and feeds them into Hailuo to generate the video.
I used Mirage for avatar generation and MiniMax for TTS generation.
Programmatic 3D After Effects-Style Video Generation
A program for generating complete 3D After Effects-style videos in code: creative planning, animated typography, 3D asset sourcing and selection, and After Effects-style null-stacked camera movements.
The agent pipeline takes in an audio file, then finds 3D assets on Sketchfab, uses a VLLM to orient the asset properly, determines typography style and camera motion, and generates the full video.
High-End Programmatic 3D Motion Graphics
Used Three.js and GSAP to create 3D motion graphics. Early explorations into if/how you could build something with LLMs that fully replaced a skilled After Effects editor.
FounderTube
I have a playlist of about 700 videos from different founders and operators I've collected on YouTube over the years. A lot of the best ones are these random 25-minute interviews or talks from 2014 with 3,000 views. So I built this to automatically surface those types of videos.
For the home feed, I created a list of 30–50 of the most impressive founders and companies, then a list of queries that would find good videos from them. The home feed randomly combines those two, sets the minimum duration and date filters, then shows you the videos.
For searching, you can filter by year range, before a certain year, exact duration range, etc.
Contact
The best way to reach me is by email.
- Work: thomas@rayloom.ai
- Personal: lancer.thomas123@gmail.com