Friends+Family DirectAI Monthly Update (June 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 our first recurring revenue contract! We’re helping a public safety company track objects over live video. We enabled them to demo object tracking on up to three simultaneous live streams at a large conference in mid-June. A few technical feats that happened in pursuit of this goal:
Transitioned to native webrtc live streams & built a custom task manager to bring end-to-end latency down to 700 ms
Switched to a different gpu, bringing unit costs down by a factor of 3
We received dozens of api signups and were packed with calendly onboarding requests after our public release in early June. We’re thrilled about the potential use cases - it’s been a blast working through the data that prospective clients have sent us.
We demonstrated near-linear scaling (both horizontally and vertically) when running a throughput test for a potential customer. Isaac completed a 12 hour test where we satisfied 6 million classification requests with 0 drops and steady 200ms latency.
What We Learned
I received some feedback from our last issue that it would be helpful to directly address learnings/failures. I’ll try to be explicit this month!
There were 2.5 months between the handshake agreement and the signed contract with our customer. That length of time (and $ we paid our lawyers) isn’t sustainable. At this stage, the greatest risk to the business is not having a signed contract rather than having a poorly negotiated contract. Credit to Ali Partovi for calling us out for this.
Learning: Once the customer is satisfied with a contract draft, time box getting the contract signed.
We failed to close a deal last week after having (what we thought was) a positive discussion with a potential customer. They described a few specific needs that we met ahead of schedule but they ended up asking to delay discussions by several weeks. Less obvious learnings here but I have some hypotheses.
Hypotheses:
Spend as long as possible diagnosing customer needs. Don’t immediately start building after pattern matching for problems your tech solves.
No deal is done until it’s signed.
Isaac and I were working in different locations for the majority of the month. Remote is useful but this reassures me that I’m not (currently) interested in running a remote company.
Learning: My best work is done with in-person teams.
What’s Up Next
We’re going to publicly release our object detection API and issue credentials to everyone who’s signed up! We’re hoping this allows leads to test/qualify themselves and surfaces a bunch of useful feedback!
We’re going to announce our integration with LiveKit via a blog post in the near future! Any application that uses LiveKit to support their live video will be able to natively integrate object tracking. Check out this draft loom we made in preparation for that announcement.
Asks
Please engage with our posts when we publish our product announcement(s)! We sincerely appreciate comments, retweets, and likes.
My calendly is here if anyone would like to chat!
Kindly,
Ben
P.S. Check out our brand new desktop!
