SubFlick AI
A cloud-native SaaS platform designed to automate subtitle generation and translation without requiring heavy local hardware. Built on a serverless architecture, it leverages the Groq Cloud API (Whisper Large v3) for near-instant transcription and Google Gemini 1.5 Flash for context-aware translations.
Key Features
- Powered by Groq's LPU inference engine for speech-to-text (approx. 30x faster than real-time).
- Context-aware translation into 60+ languages (Sinhala, French, Spanish, etc.) via Google Gemini.
- Auto-generation of valid .SRT files with precise timestamp alignment.
- Bandwidth-optimized architecture with client/server-side audio extraction.
- Zero-GPU requirement; runs entirely on free-tier cloud infrastructure.
Technical Challenge
The primary hurdle was migrating from local inference to a serverless cloud architecture, which exposed strict API payload limitations. I initially faced errors when injecting raw transcript strings directly into the Gemini API due to token constraints. I resolved this by re-architecting the pipeline to handle transcripts as file-based inputs (.SRT) rather than raw text. This not only bypassed the character limitations but also ensured that timestamp formatting remained intact during the translation process.
"Focusing on system optimization and clean data patterns."