SpoGoRun Agent

Architecture of the Project
Architecture in ADK UI

inspiration

i love running outdoors. but somewhere along the way, my brain gets too fixated on exhaustion, and i end up stopping way earlier than i should. one day, while running with spotify on, i caught myself playing a little game: trying to guess the song from just the first few notes of the intro. and something clicked. that tiny mental challenge was enough to pull my attention away from tired legs and heavy breathing. i ran further, felt better, and actually had fun doing it. that moment stuck with me. what if this experience was properly gamified? what if an ai agent could turn every run into a music quiz, tailored to your taste, hands free, no screen needed? that's where spogorun was born.

what it does

spogorun is a voice-first ai agent that turns your run into a music game. once you start your run and launch the agent, you can speak naturally to play songs, guess what's currently playing, or kick off a full song-guessing quiz based on your favorite genre, artist, mood, or category. no tapping, no scrolling. just you, your voice, your music, and the game.

how it was built

the first step was proving the idea was even possible. i prototyped it using coding agents and an open source project called spogo, a cli tool for interacting with your spotify account. it worked well enough to validate the concept. from there, the goal was to give it a real agentic backbone. i integrated the spogo cli with the adk gemini live toolkit to orchestrate spotify through speech. the early features were simple: play a song, check device status, add to queue. then came the fun part. i built out more interactive flows: guessing the currently playing song, creating smart song lists from natural language prompts, and finally, a full quiz mode where the agent builds a 5-song list, plays each track one by one, and gives you two guesses per song. once everything was tested and stable, i deployed the whole thing on cloud run on google cloud.

challenges ran into

the first real challenge was getting up to speed with the adk and gemini live integration. understanding how the toolkit was architected made a huge difference in how i thought about the overall system and the role of each component. the second challenge was weaving the spogo cli into a multi-step, stateful game experience. it required digging deep into spotify's authentication specifics and building efficient, reliable tools that the runner and root models could use in sync with the overall game flow.

accomplishments proud of

getting the first prototype working was a big moment. it confirmed the idea was real and buildable, and that confidence carried everything forward. the part i'm most proud of is how naturally the agent handles the nuance of a user's guess versus a user humming or singing a song. that distinction made the ux feel genuinely smooth and smart. this kind of integrated, preference-aware, screen-free song guessing experience is something that siri, standalone llms, and even spotify don't offer today.

what i learned

building something you actually love using is both incredibly fun and a lot of work. going from idea to alpha taught me just how many layers there are between a concept and a real product. on the technical side, the adk agents framework was a joy to work with. the toolkit architecture made it much faster to build and iterate, and kept the focus where it should be: on the product experience. (and personally, this project has me more motivated to go outdoors for a run during this spring break!)

what's next for spogorun

spogorun is the first ai-native experience that brings running, music, and fun together in one seamless flow. no screens, no buttons. just a conversation that adapts to your preferences from the moment you start moving. the next step is wrapping it in a simple, nostalgic, coverflow-inspired mobile app. the goal is to make the whole experience feel as good to look at as it does to use.