At FastPix Inc., I designed and implemented a multimodal video and audio moderation pipeline using Gemini on Vertex AI. It produces timestamped decisions that support automated flagging, blurring, muting, beeping, cutting, or human review.
I added multilingual transcription with word-level timestamps and repeated-occurrence detection, including support for Indic languages. Context-aware rules and prompt engineering cover 11 policy categories across real, fictional, cinematic, news, and interview content.
I also built document-processing pipelines for scanning, OCR, and information extraction, replacing manual data entry with automation. For the AaoNXT OTT platform, I integrated FastPix Video SDKs and migrated playback across mobile, Android TV, and Chromecast, supporting DRM and HLS streaming.
In my Scrivara project, I built a self-hosted document automation pipeline that combines PDF text extraction, OCR, and vision-language models, with confidence-gated field mapping. I also trained a neural network to classify more than 10 cyberattack types and deployed real-time inference through a Flask web interface.
