Mustafa Erhan Portakal
Türkçe

Auraly-AI

Auraly-AI is a mood-based music recommendation product built by Mustafa Erhan Portakal at Vimof Studios, which interprets a listener's mood with Google Gemini and turns it into a playable Spotify playlist. It runs as a TypeScript monorepo of four applications — a NestJS API, a Next.js web app, an admin panel and a React Native mobile app — and is live at auraly-ai.com.

What Auraly-AI does

Auraly-AI takes a description of how someone feels and returns music they can play immediately. Google Gemini interprets the mood and generates a playlist concept; the Spotify Web API resolves it into real tracks and plays them on the listener's own Spotify session. It is the product behind Vimof Studios, the two-person studio I co-founded in July 2025, and it was also my graduation project at Bursa Technical University.

How the system is put together

The whole product lives in one TypeScript monorepo containing four applications that share a single typed DTO package, so a change to an API contract is a compile error in the web app and the mobile app rather than a runtime surprise.

  • API — NestJS 11, with 102 REST endpoints across 17 controllers and 27 MongoDB schemas modelled in Mongoose 8.
  • Web — Next.js 15.
  • Admin panel — an internal application, kept off the public internet behind Cloudflare Access.
  • Mobile — React Native on Expo, routed with Expo Router, covering audio playback, push notifications, deep linking and localisation through i18next and expo-localization.

Two external services carry the product's core behaviour. Spotify handles identity and playback: OAuth, token refresh, library access and Connect playback control. Gemini handles interpretation: reading a mood and generating a playlist.

Deployment is Docker Compose behind Nginx, with CI/CD on GitHub Actions.

Why the recommendation pipeline had to be rewritten

Spotify deprecated its Recommendations endpoint, which removed the single API call the entire feature had been built on. Recommendation was not a side feature of Auraly-AI — it was the product — so this was not something that could be patched around.

I rebuilt the pipeline rather than looking for a drop-in replacement. The new algorithm was designed and documented as a deliberate piece of work, then put behind a feature flag with variant positioning so the replacement could be introduced and compared without a hard cutover. Writing the design down first mattered here: the constraint was permanent and external, so the reasoning needed to survive past the moment of the outage.

What is still open

Playback synchronisation between Spotify and YouTube is in progress. Keeping a single listening session coherent across two providers with different playback models and different notions of session state is the current problem.

Where it stands

Auraly-AI is live at auraly-ai.com, with 241 commits over 13 months of development. The iOS and Android release is targeted for August 2026. I handle development; my partner at Vimof Studios handles QA and testing.