Friends+Family DirectAI Monthly Update (May 2023)
Hey all!
Keeping you up to date with what’s new at DirectAI. To reiterate, we’re enabling anyone to build computer vision models that classify, detect, and track objects in images & videos, without training data. If you missed last month’s update, check it out here!
Last Month’s Rundown
We received our first $ of revenue on an image labeling product! We drew bounding boxes for a company doing object detection inside of warehouses. Labeling quality wasn’t great so we’ve decided to table this market for the time being.
We ran a public release on Saturday! You can still engage with our twitter thread, check out our Product Hunt post, or play with the detection demo on our fancy new landing page.
Some hypothesis invalidation:
We pitched to 16 packaging manufacturers at LuxePack on automating their Quality Control process. Lots of teams are interested in “defect detection” but the root problem is an inability to “do something” about defects. We’d need to automate the process of removing defects from the assembly line to really unlock value.
We talked to a company interested in counting boxes on pallets in warehouse staging areas. We were unable to meaningfully help them due to box obstruction in images.
Some product improvements:
Built an annotation tool & accompanying frontend interface in a week
Website & Demo re-vamp
Our models work better in low-framerate video environments & our model configuration is more powerful/controllable
Added support for our customer’s new livestream video service
What’s Up Next
Engineering Work:
Transition our livestream support off a more inefficient protocol (RTMP -> WebRTC).
Transition to a better GPU-scaling software.
Get our software ready to be run on our customer’s servers.
Productionize our API usage-tracking and billing.
If we have time -> add automatic model improvements according to live user feedback.
Big Goal -> Plug & Play sub-500ms end-to-end latency @ 30fps
We’re collecting more data on our hypothesis that companies outsourcing video/vision infrastructure are more receptive to outsourcing AI. Our current customer falls in this bucket, but more importantly, our customer’s video provider is actively searching for a platform ML solution. We’re talking with a robotics hardware provider about a similar approach. TLDR: We want to partner with infrastructure services to be the de-facto ML provider on their platform.
Asks
We’d like feedback on our demo and our launch strategy! Please tell us what left you confused/what doesn’t work well. Even better if you join our Discord and post it publicly!
My calendly is here if anyone would like to chat!
Kindly,
Ben
P.S. Check out the conference we went to!