Colin Mitts
Mathematician · Builder · Strategist
I build production AI systems across video, language, and agentic workflows.
I've been chasing AI since 2011. A failed computer-vision startup taught me two expensive lessons: if you don't ask the right question you build the wrong thing, and if you can't build it yourself, you're stuck waiting on someone who can. So I got an MBA, learned to code for real, fell for NLP at an Intel-acquired startup, and have been building language systems ever since. The last two years: multimodal and agentic systems.
Off hours: piano, skiing, cooking, proud new father.
video · multimodal
Video Intelligence
Give it a video, get back a report that explains it — the storyline in beats, every scene with what was said and what the picture shows. The demo does the talking.
documents · text + vision
Order Form Extraction
Hybrid text-and-vision extraction that reads thousands of scanned paper order forms into a clean, searchable database — 5,800+ line items, with a second-pass vision check on the fields most expensive to get wrong.
imessage · agentic · evals
iMessage Baby Tracker
Built for my infant son. Caregivers text the family group chat like normal humans — an LLM pipeline turns the chatter into structured care analytics, daily goals, trends, and twice-daily briefings back to the chat.