Friends+Family DirectAI Monthly Update (December 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 signed a new customer - we’ll be doing automated video editing for a team that produces customized shot analysis videos for golfers.
We agreed on pricing for a third project and drafted our first contract with an industrial camera company - we’ll be doing on-prem anomaly detection on live video for them.
We ran a large-scale batch inference (> 500k images) for an existing customer.
We mentioned last month that we were planning to do small but frequent product releases. We built some interesting internal tools but failed to prioritize a public release.
What We Learned
On Public Releases:
There’s evidence of latent desire for access to our livestream video analytics endpoints. What’s more, we haven’t released an endpoint publicly since our initial launch in June 2023. While we’ve been building all kinds of 1:1 solutions for customers privately, it seems like now is a good time to start a cadence of feature releases.
On Onboarding Flow:
Our onboarding flow is starting to be a bottleneck. We’re currently emailing client credentials (with the secret contained in a one-time view link). It’s secure but not a great user experience. As an example, one user at a partner company had to ask us to regenerate their credentials because the secret link expired. See attached:
On Prioritizing Model Improvements:
We’ve increasingly been “hacking together” elements of our existing tech to construct pipelines for prospective customers (i.e. What is the curve of best fit for a moving object? Are two objects about to collide? At what moment did action X occur in a video?). We’re wondering at what point it makes sense to invest time in training a custom model. until now we've been adding lots of proprietary math to existing open source models. An upfront training cost would make these complex video analytics much simpler to build.
What’s Up Next
Our big goal this month is to sign a contract with and ship a complete product to the industrial camera client.
Outside of that, we’d like to publicly release four pieces of DirectAI software for users to play with. That necessitates us improving our authentication flow by 1) migrating to a new client credentials manager and 2) making credentials web-app accessible. We also need to start polishing internal tools so they’re ready for external consumption!
Asks
Please be on the lookout for our releases this month via LinkedIn & Twitter! We’d appreciate it if you gave DirectAI’s features a test run and liked/shared/commented on the posts!
As a friendly reminder, our shared calendly is here if you’d ever like to get in touch! And if you have feedback on the update, feel free to reply directly!
Kindly,
Ben
P.S. Check out this live re-broadcast demo that Isaac built last month!