THE THING WE BUILT:

Scout

Finding new music used to be simple. You’d ask the cool record store employee with obscure band patches on their jacket. Now algorithms feed us an endless loop of what we already like, turning discovery into an echo chamber.

Sure, you could ask ChatGPT for recommendations. But it’ll suggest bands everyone already knows.

And music streaming services? They’re optimized to push whatever keeps you on the platform, not what might genuinely blow your mind.

So we built a thing.

Scout in action

How it works:

• Enter the name of a band you like (try “This Will Destroy You” if you need inspiration)

• Scout searches Reddit for human conversations about similar music

• AI extracts band names from those authentic recommendations

• Scout creates a YouTube playlist of these bands

• Discover your next favorite band based on actual humans, not algorithms

Things we learned from the thing:

1. Multiple sources beat a single LLM. When we first tried using just LLMs for recommendations, the results were either too generic or painfully obvious. Real Redditors suggest more interesting bands than any algorithm. The magic is when AI helps process human recommendations instead of generating its own.

2. LangChain is great at orchestration. Building this app meant connecting several services (Reddit via SerpAPI, OpenAI, Tavily for YouTube links, and our Next.js + React frontend). LangChain was great at connecting this multi-step pipeline. It's becoming our go-to for complex AI-centered workflows.

3. But Regex isn’t great at natural language. We started with simple string matching to extract band names from Reddit comments… but it failed spectacularly. Human music conversations are messy. “You might dig The End Of The Ocean” is easy for humans to understand but tricky for Regex-based extraction. Thankfully, LLMs understand context in ways Regex can’t.

4. Human Curation + AI Processing = Novel Suggestions. The 🪄 happens when these pieces work in sync—curation, extraction, delivery. Niche Redditors discussing Godspeed You! Black Emperor or Russian Circles know more about post-rock than any algorithm or LLM we’ve tried. Meanwhile, AI extracts and processes this knowledge, and YouTube pairs it with actual music videos. Alone, each is incomplete. But combined, they create something that feels almost like asking that record store employee for recommendations. Just with broader reach and no judgment about your taste in bands (well, we’ll still judge).

Built with ❤️ by Township

You’re receiving this because you signed up for Township’s “We Built a Thing” newsletter. We'll only send these when we, well, build things.