Ahmed Umair Khan

Software Engineer

Hyderabad, Telangana, India6 yrs 10 mos experience

AI EnabledAI ML Practitioner

Key Highlights

Expert in building large-scale data processing pipelines.
Significant contributions to ASR and TTS technologies.
Proven track record in optimizing machine learning models.

Stackforce AI infers this person is a Backend-heavy Fullstack Engineer specializing in AI and Machine Learning technologies.

Contact

Skills

Core Skills

Machine LearningArtificial Intelligence (ai)Data PipelinesSpeech RecognitionData ProcessingData Analysis

Other Skills

Python (Programming Language)PySparkMLOpsLarge Language Models (LLM)FastAPIKubernetesDockerData ScienceGenAILangChainLanggraphRetrieval-Augmented Generation (RAG)AI AgentsNatural Language Processing (NLP)Kubeflow

About

System Software Engineer with experience at NVIDIA, specializing in building and optimizing large-scale data processing pipelines, machine learning models, AI and GenAI applications using LLM. Experienced in ASR, TTS, and translation pipelines, with a background in developing robust tools and automated systems to enhance model development and evaluation efficiency.

Experience

6 yrs 10 mos

Total Experience

6 yrs 10 mos

Average Tenure

6 yrs 10 mos

Current Experience

Nvidia

System Software Engineer AI ML

Jul 2019 – Present · 6 yrs 10 mos · Hyderabad, Telangana, India · Remote

TTS Data Pipeline (Kratos): Engineered an automated Text-to-Speech (TTS) data pipeline on Kratos for filtering, conversion, and segmentation. Developed a dashboard to visualize stats and track loss. Implemented a catalog-backed versioning system for data reuse.
Arabic ASR Model (Maglev): Developed an end-to-end pipeline for Arabic ASR datasets. Optimized evaluation by adding a normalization step, which reduced the Word Error Rate (WER) by 3-4% and confirmed transcription correctness.
PDF Metadata Extraction (Kratos): Created a Python module and an efficient parallel pipeline to extract metadata from files in SwiftStack. Developed a specialized extractor for PDFs, supporting both text-based and OCR-based extraction.
Speech Data Processing (Kratos): Engineered an end-to-end pipeline that automated ASR inference and Diarization. This processed over 268 TB of English audio data for an LLM model training pipeline.
Data Analysis (Scrapy Cluster): Engineered a PySpark data analysis pipeline to process over 10 million crawled URLs. Provided data-driven insights to define and implement an effective stopping criteria for the crawler.
Synthetic Data Generation (SLURM): Optimized a Hindi translation pipeline by integrating Nvidia's Crossfit (doubling throughput) and CTranslate2 (a 10-12x performance boost). This resulted in a successful open-source contribution.
ASR Pipeline Features & Tools: Automated the creation of a development (DEV) set (converting a manual process). Implemented a metric stage and developed a unified evaluator tool to standardize ASR evaluation across various Cloud Service Providers (CSPs).
ASR Datasets Creation: Engineered and built large-scale ASR datasets for both Russian and Arabic. Executed a comprehensive data pipeline covering ingestion, normalization, ASR inference, segmentation, metric calculation, and curation.