Building a Meeting-Matching Tool That Doesn't Feel Creepy

Every NOG event I build for has the same attendee list problem. A long list of names, sorted alphabetically, maybe a company column if you're lucky. You scroll it once on the train there, forget it exists by the time you arrive, and spend the event talking to whoever happens to be standing near the coffee at the same moment you are. Which is fine. Some of the best conversations at these things start exactly that way. But it also means the engineer from a transit network who's been trying to reach a couple of ISPs in the room for a while never finds out they're standing across the room.

So a while back I added a feature to NOGnet that looks at what an attendee's company does and whatever goals they've typed in for that specific event, then quietly suggests a handful of people worth finding, tucked under a header that just says "people you might want to meet."

Building the matching logic was the easy part. Comparing a couple of short text fields and ranking candidates isn't hard. I had a rough version working in an afternoon. What took actual thought was everything around it, because this crowd has a strong allergy to being sold to, matched, or funneled. These are people who spend their working hours keeping traffic flowing between networks that don't fully trust each other. They come to these events for the hallway track, and for the argument about routing policy that starts at the bar and doesn't really end. A tool that announces "here is your optimal networking path" is dead on arrival. Worse, it earns active resentment on the way down.

What I ended up stripping out

The first version showed a match percentage next to each suggested name. I killed it after a day of looking at it myself. Seeing a number next to a real person's name felt like something out of a dating app that wandered into the wrong conference. It turned a genuinely useful suggestion into something that felt done to the attendee, rather than offered to them.

I also pulled back on explaining the reasoning behind each match. My instinct as an engineer is to show my work: here's why we think you two should talk, here's the overlap. But spelling that out too precisely slides into surveillance territory fast, and nobody wants a paragraph explaining how the system inferred their priorities from three form fields. The current version says something short and plain, like "works in transit, similar region", enough to make the suggestion feel earned without reading like a dossier.

The suggestions only show up if someone goes looking for them. Nothing pushes a notification mid-conversation telling you to go talk to someone else instead. You open the tool, or you don't, and the event carries on either way.

The tension I haven't fully resolved is that the more useful this feature gets, the more it risks replacing the exact kind of unplanned encounter that makes these events worth flying to in the first place. A recommendation that's too good at finding you the right person quietly narrows the room down to a checklist. I don't think a ranked suggestion should ever outcompete the accident of standing next to the right stranger at the espresso machine. I've tried to build it so it can't, mostly by keeping it boring and easy to ignore.

Running NOGnet solo means every one of those calls is mine, and I'll get some of them wrong before organizers tell me which. For now the rule I keep coming back to is simple: suggest quietly, and never let someone feel like they've been computed.