Orbit

Three days. Hundreds of sessions and people. One connection you can’t afford to miss. Orbit finds who and what matters to you + where to catch them—while showing organizers what attendees came for.

Comment

Inspiration

API World is my first-ever tech conference, and I quickly realized that getting here was the easy part.

The hard part was figuring out where to go, what to attend, and who to meet.

189 sessions. 165 speakers. Three days. I could scroll the agenda, search LinkedIn, Google speakers, or ask ChatGPT, but I was still piecing everything together myself.

So I started with a question:

What if the conference already knew what I came here for?

That's Orbit.

Then I noticed something else: every attendee arrives with something they want to learn, solve, build, or discover, but organizers rarely get to see what their audience is actually searching for.

Orbit helps attendees find what they came for, and helps organizers understand what they came for in the first place.

What it does

Orbit starts with one question:

What are you trying to get out of this conference?

Tell it what you're building, what you're stuck on, or what you want to learn. Orbit searches across the conference and recommends:

People to meet, ranked by relevance Sessions to attend, based on the actual problem rather than keywords Why they matter, so every recommendation is explainable Where to find them, connected directly to the conference schedule

But Orbit also works at the aggregate level.

Each query becomes an anonymous signal about attendee intent. The organizer dashboard surfaces what people are trying to solve, which sessions and speakers are meeting those needs, and, perhaps most importantly, what attendees are asking for that the program doesn't answer well.

The attendee gets a better conference. The organizer gets a better understanding of their audience.

How I built it

Orbit starts with the conference itself.

Using the Sessionize API, I ingest API World's program into a structured corpus of 189 sessions and 165 speakers, including schedules, rooms, categories, profiles, and speaker-session relationships.

But conference data is designed to populate an agenda, not answer semantic questions. Many entities have little more than a title, name, or company.

So I built an enrichment pipeline using SerpApi to add relevant public context to sparse speakers and sessions. Results are cleaned, structured, and cached so the expensive discovery work only happens once.

I then create embeddings over the enriched corpus.

When someone asks Orbit a question, it:

Understands the attendee's intent Retrieves and ranks relevant speakers and sessions Grounds the response in real conference entities Explains why each recommendation matters Connects recommendations to actual rooms and times

At the same time, Orbit captures aggregate signals about what was asked, what was recommended, and where no strong match existed.

One query becomes both a recommendation for the attendee and a signal for the organizer.

Challenges I ran into

  1. The agenda wasn't enough

A conference API can be perfectly structured and still contain very little meaning. Sparse titles and profiles weren't enough for reliable semantic retrieval.

That led to Orbit's enrichment-and-cache architecture: enrich the corpus once, then reuse that knowledge across attendee queries.

  1. People don't ask questions like agendas are written

Attendees describe problems. Conference programs describe topics.

Someone might ask:

'My agent keeps making expensive calls and I don't know where the tokens are going.'

The relevant session may be about AI observability or token optimization without using the same language.

Keyword search wasn't enough. Orbit needed to retrieve based on meaning.

  1. The closest result isn't always a good result

Semantic search will almost always return something. That doesn't mean it should be recommended.

But that led to another insight: a failed recommendation isn't necessarily failed data.

If attendees repeatedly ask about something the conference can't answer well, that gap itself becomes useful information for organizers.

Accomplishments that I am proud of

The thing I'm most proud of isn't that Orbit can search 354 conference entities.

It's that I'm testing it inside the conference it was built for.

I put Orbit in front of API World attendees and tracked what they asked, what it recommended, where it struggled, and whether they would actually act on those recommendations.

Watching real attendees enter problems I never anticipated turned Orbit from a solution to my own frustration into something I could test with actual users.

What I learned

Recommendations aren't the product. Decisions are.

Attendees already have agendas, Google, LinkedIn, and ChatGPT. What they need is help answering:

Given why I'm here, where should I spend my next hour?

I also learned that the questions can be as valuable as the answers.

Successful matches reveal what the program serves well. Failed matches reveal what might be missing.

That's what shifted Orbit in my mind from an AI conference recommendation tool toward a conference intelligence layer.

What's next for Orbit

The next step is to take Orbit beyond a single conference.

I want to test whether the signals it captures can help organizers adapt while an event is still happening, surfacing unmet attendee needs, identifying valuable connections, and revealing where the program has gaps.

Longer term, Orbit could become a reusable intelligence layer across conferences: helping attendees make better decisions in the moment, while giving organizers a continuously improving picture of what their communities actually need.

The goal isn't to help people search a conference. It's to help conferences understand the people inside them.

Built With

Share this project:

Updates

Submission history