Skip to content
NNadeem Shabir
Back

Experience

Data Engineering Intern

Quansys AI Technologies Pvt. Ltd., January 2026 to April 2026

Fine tuned Whisper on a cleaned Hindi corpus, and built ingestion that can resume after a crash.

  • Whisper
  • A100
  • speech data
  • data pipelines
01

Fine tuning

  • Fine tuned OpenAI Whisper on NVIDIA A100 GPUs.
  • The corpus was a custom cleaned Hindi speech set of more than 27,000 files.
  • Work covered both the data preparation and the training run.
02

Preparation

  • Schema transformation, deduplication, text cleaning, and timestamp aligned audio chunking.
  • The output was one training ready format.
03

Ingestion

  • A crash resumable, VM safe pipeline for 35+ Indic speech datasets across 20+ Indian languages.
  • More than 10 TB of audio from Hugging Face and Mozilla Common Voice.
  • Memory efficient extraction, so a long job does not repeat work it has already finished.