This is a submission for the MLH x DEV Writing Challenge
What I Built
I built Throughline, a real-time lecture companion, solo at SBUHacks 2026 (Stony Brook University's hackathon).
Lectures don't happen in isolation. A professor says "remember what we did with this last week?" and half the room has no idea what "this" refers to. I take most of my classes in English, which isn't my first language, so I'd already been recording and transcribing lectures for myself. Throughline came from asking: what if those transcripts could talk back during class?
Here's how it works:
- Live alignment: While the lecture is happening, Throughline transcribes what's being said and, every 2 seconds, links it to the most relevant parts of previous lecture transcripts and slides. When the professor references an old concept, the original explanation is already on your screen.
- Explain & summarize (Claude): Tap any moment to get a plain-language explanation or a running summary of the lecture so far.
- Semantic search (Gemini): Search across every past lecture by meaning, not keywords. "When did we talk about why the stack grows down?" works even if those exact words were never said.
- Post-class oral exam: After class, Throughline asks you questions out loud about what was just covered and evaluates your spoken answers, so you find out what you didn't actually understand before the exam does.
What I learned
I didn't win. The projects that placed were simpler, and honestly less technically ambitious, but they were easy to understand in three seconds and fun to watch live.
That stung, but it taught me something I needed to hear: in a hackathon, you're not judged on what you built, you're judged on what the judges understood in a few minutes. I spent my time building four features. I should have built one, and spent the rest of the time making it impossible to miss on stage. Real-time alignment is the core of Throughline, but in my demo it was just one of four things on the screen.
I'm still using Throughline for my own classes, and I'm carrying that lesson straight into my next hackathon.
Demo
https://youtu.be/_sP9_kvxxkY?si=R-HAn4H_p5vj5VOj
throughline.ideantoe.com
Partner Technologies
- Google Gemini: Powers semantic search across all past lectures. [How you used it — embeddings model name, how results are ranked]
- Anthropic Claude: Generates explanations and summaries of the live lecture.
Top comments (0)