Hi, my name is

Narain.

I create intelligent systems and applications

I am a passionate Software Engineer and cloud developer. I use modern technologies to build smart, efficient and user-friendly applications.

About Me

I am a Software Engineer. My work focuses on levaraging advanced AI technologies, cloud platforms and performance optimization techniques to build intelligent, scalable applications. I am passionate about the intersection of technology and innovation, striving to create solutions that make significant impact. Here are a few technologies I've been working with recently:
  • AWS
  • Azure
  • C/CPP
  • CUDA
  • Javascript
  • Python
  • PyTorch
  • Spark

Experience

MLOps Engineer (Part Time) - ASU Enterprise Technology
Jan 2024 - present

– Researched and Developed vector databases and data engineering pipelines for scalable data retrieval and question answering tools, utilizing ASU’s documents to improve accessibility and support decision-making in administrative and academic functions.

– developed an embedding model for academic and research documents using masked language modeling for fill-in-the-middle and next token prediction. The model utilizes SentencePiece embedding and the DeBERTa architecture to enhance semantic understanding of complex texts.

– Implemented robust, scalable solutions for machine learning model deployments and maintenance, reducing operational delays and improving system reliability and query precision.

Software Developer - Fidelity Investments
April 2022 - Jul 2023

– Developed and implemented robust infrastructure and pipelines for the deployment of machine learning models on AWS.

– Engineered online feature stores to support real-time data and optimized batch inferencing to handle large-scale data efficiently.

– Contributed to the deployment of diverse machine learning models, including document layout models, YOLO for real-time object detection, and large language models, further expanding the scope and impact of deployment solutions.

– Led the deployment and performance optimization of large language models, including BLOOM and Flan-T5, using tensor and model parallelism to accelerate token generation and enhance model performance.

AI Researcher - QPiAI Technologies
May 2021 - Apr 2022

– Developed an AI-driven tool for extracting circuit diagrams as a graph of connected components from documents.

– Implemented a graph convolutional network (GCN) to train graph contrastive learning models for different circuit layouts, achieving high recall in identifying similarities between different circuit layouts.

– Introduced a custom ranking algorithm for product recommendations supporting automated price estimation for sales personnel.

– Developed vision algorithms for safety monitoring systems in warehouses using cameras to ensure compliance with social distancing.

Data Analyst - Thoughtware Analytics
Jan 2020 - May 2021

– Conducted demand forecasting for manufacturing and service-oriented businesses, focusing on supply chain improvements.

– Designed and Executed computer vision algorithms for quality assurance in manufacturing processes, reducing material waste and increasing production quality.

Education

2023 - 2025
MS Computer Engineering
Arizona State university

Courses Taken

  • Algorithms, Digital image processing, Data intensive systems for machine learning.
2016 - 2020
Bachelor of Technology in Mechanical Engineering
Amrita Vishwa vidyapeetham

Projects

Whisper Training on Indian languages
Cuda Deep Learning NLP
Whisper Training on Indian languages
Enhancement of Whisper Model for South Asian Language Speech Recognition, I focused on refining the Whisper v3 medium model to significantly improve speech recognition accuracy for South Asian languages.
Steganalysis Detection
Python Deep Learning computer vision
Steganalysis Detection
Developed a CNN-based model to detect steganographic content within digital images. Modified the traditional ResNet architecture by integrating Efficient Channel Attention (ECA) and reducing the stride of convolution to enhance the model's sensitivity to minute, concealed data.
Indic LLama
WASM LLM Deep Learning
Indic LLama
I enhanced a LLaMA model to improve its performance across multiple languages and domains by integrating a newly developed tokenizer and training it on a diverse set of translated instruction datasets and domain-specific dialogues.
LLM Model inference in the browser
Javascript Deep Learning WebAssembly C GPU
LLM Model inference in the browser
This project enables to inference open sourced large language models like gemma, LLama and mistral in the browser with zero installation required. All compute happens on the clients PC using webassembly SIMD.
Compiler System for Deep Learning Graph Optimization and Op-Fusion
TVM TASO ARM NEON CUDA Jetson Orin
Compiler System for Deep Learning Graph Optimization and Op-Fusion
Developed a compiler system for mobile backends using TVM and CUDA, achieving performance boosts. Designed SIMD kernels for ARM NEON and custom CUDA kernels for Jetson Orin, outperforming OpenCV, torchscript benchmarks.

Get in Touch

My inbox is always open. Whether you have a question or just want to say hi, I’ll try my best to get back to you!