SongScript
A language learning app focused on song transliteration. Users can learn pronunciation of songs in foreign languages through phonetic transliteration. Features a WhisperX pipeline for audio processing with word-level timestamps, real-time sync between lyrics and audio, and progressive difficulty levels. Supports Persian, Korean, Arabic, Hebrew, and Spanish with Exa-powered lyrics discovery.
Built with
Python
The journey
01
The Idea
Learning songs in foreign languages means learning pronunciation, not just reading lyrics. Existing apps show text but don't bridge the gap between written lyrics and how they actually sound. SongScript was born from wanting to sing along to Persian, Korean, and Arabic music.
02
WhisperX Pipeline
Built a Python audio processing pipeline using WhisperX for word-level timestamps: download audio → separate vocals → transcribe with language-specific models → generate transliteration. Each word gets a precise start/end timestamp for real-time sync.
03
TanStack + Convex Stack
Chose TanStack Start for the frontend and Convex for the backend — reactive database with real-time sync, no REST API boilerplate. Songs, lyrics, words, and user progress all live in Convex with typed schemas.
04
Exa Song Discovery
Added a zero-cost lyrics pipeline using Exa search: 3 searches per song get lyrics + transliteration + English translation for popular songs. Claude generates word-by-word breakdowns at zero API cost. Tested with Persian Bandari, Spanish, and Korean songs.
05
Multi-Language Support
Expanded beyond the initial Persian focus to support Korean, Arabic, Hebrew, and Spanish. Each language has its own alignment model for WhisperX and Unicode-aware punctuation handling. 199 tests across the pipeline.
More projects
Cantaloupe AI - Recruitment Platform
Enterprise-grade AI-powered recruitment platform with voice interviews, resume screening, multi-channel communication, and seamless ATS/HRIS integrations. Automates hiring workflows for Quick Service Restaurants, healthcare, and construction industries.
Hand Sign Detection Model
A production-ready computer vision model that detects and classifies hands, arms, and non-hand objects in real-time with 96% accuracy.
BrainLayer — Persistent Memory for AI Agents
12 MCP tools, 335K+ indexed chunks, hybrid semantic+keyword search with MMR dedup, knowledge graph with entity resolution, Gemini enrichment. 1,848 Python + 54 Swift tests. pip install brainlayer.