Daniel Ruppin

A full-stack playlist-triage dashboard that streams a user's Spotify library through OAuth, plays tracks with the Web Playback SDK, and files them into destination playlists in real time.

• Aggregates Spotify metadata, Last.fm tags, and ReccoBeats audio features into one per-track context object, replacing Spotify's deprecated Audio Features API.
• Provider-agnostic LLM layer (Anthropic, OpenAI, Google, Groq, OpenRouter) generates grounded subgenre/mood tags, with prompt-hash caching to cut redundant API calls.
• Encrypted session management for OAuth tokens, plus a responsive three-panel React UI with optimistic state and multi-destination filing.

Marketing site for a Milwaukee-area handyman business, built and deployed end to end — design, front end, and a small Worker-backed contact form.

• Static site served from Cloudflare Workers with a Worker script handling the contact-form API.
• Homepage and full photo gallery driven by a shared data file, so adding a job photo is a one-line change.
• Deployed to a custom domain with production and www routing.

A Python evaluation pipeline measuring the accuracy and robustness of an AI music-transcription model used in a web-based student practice tool.

• Automated large-scale experiments with Slurm batch jobs and JSON configs, aggregating results across datasets and noise conditions with pandas.
• Applied statistical testing and documented model limitations and trade-offs to support the sponsor's decisions.

Analyzed a client's invoice and payment data to identify billing-category patterns and support a more efficient payment-classification workflow.

• Designed and evaluated baseline models for invoice filing and key-field extraction, focused on reducing mislabeled payments.
• Worked directly with the project sponsor to refine categories and translate findings into practical workflow changes.