Why now
Building got cheap. Knowing what to build didn't.
Most of what's worth building doesn't show up on anyone's list. It got normalized years ago, back when there wasn't another way to do it.
The systems large companies run on were never withheld from everyone else. They were just priced out of reach. That's the whole gap. It isn't a gap in ambition or in talent, and it never was.
What changed is the cost of building. The part that didn't get cheaper is knowing which thing is worth building, and how to land it so it still works after the person who built it walks away.
Recent work
A pipeline from supplier to storefront.
Sparktec Motorsports
A live supplier-to-retail catalog pipeline.
Before
Product data arrives the way suppliers send it — every vendor its own format, its own gaps. Some listings come with no usable description. Some are missing the attributes a listing needs outright. Images live somewhere else. Getting any of it onto the storefront in the store's own voice is a listing-by-listing job, so the catalog stays behind what's actually available to sell.
After
Listings come through in Sparktec's own voice — images pulled and matched to the right products, in a format the storefront takes directly. Where a supplier's data is too thin to work from, the pipeline hands that listing to a person instead of inventing a spec. Two vendors ran through the same pipeline. The catalog stops being a backlog and becomes something that keeps up.
Catalog work of this size normally takes a team weeks — but speed isn't the point of it. The point is that it can be re-run: new vendor, new season, changed supplier format. And that it knows where it stops — in a parts catalog, a confident wrong answer costs more than a missing one. The pipeline is running, and the next vendor is in flight.
This one happens to be auto parts. The shape of it isn't. It shows up anywhere a business already holds the information it needs but has no way to move it — from a supplier to a storefront, from a booking to a follow-up, from a finished job to the thing you'd send a past customer next.
How this works
Two ways in, depending on who should end up holding it.
Your team learns to run it
Training against your actual workflows and your actual data — not a generic course. The people who already know the business end up being the ones who operate the new capability. Nothing walks out the door when the engagement ends.
Sunrai builds it and runs it
For work that has to happen whether or not anyone is watching it. The system gets built, put into production, and operated — with someone accountable when a vendor changes a format at 2am.
Who's behind it
Ed Tang
Founder · San Francisco Bay Area
Ed spent 21 years at Google and NVIDIA. He turned up data center campuses across three continents and delivered gigawatt-scale compute, storage, and network capacity — the kind of load that powers hundreds of thousands of homes.
Most of that work was one problem wearing different clothes: grow the output faster than the headcount. Project volume grew more than tenfold without the team growing with it — by standardizing what could be standardized, automating what could be automated, and training people to run it after he left.
At NVIDIA he landed the rearchitecture that took the global network off leased carrier circuits and onto infrastructure the company controls end to end.
The scale doesn't transfer. The method does.
Find the work that's holding a business back. Build the thing that removes it. Leave the people who were already there able to run it on their own.
“I'd like my community to thrive as much as the companies I used to build for.”Ed Tang, founder
Curious whether your version of this is worth doing?
That's usually a short conversation.