Somasree Majumder

Backend Developer+Machine Learning Engineer

Download Resume
Somasree Majumder

About Me

Passionate Software Engineer and Machine Learning Engineer with a strong foundation in full-stack development and deep learning technologies.

I'm a Software Engineer at Deloitte with expertise in backend development and machine learning. With a background in Electronics and Communication Engineering, I've developed a unique perspective that combines hardware understanding with software expertise. I specialize in building scalable backend systems, implementing machine learning solutions, and developing efficient algorithms.

My experience spans across various domains including full-stack development, deep learning, and system design. I'm particularly interested in solving complex problems at scale and contributing to open-source projects. As an active member of the tech community, I've contributed to major projects including PyTorch, Hugging Face Transformers, and Keras.

Projects

Memcached server

Custom implementation of a Memcached server

Tech stack used in this project: JavaScript,Typescript

View on GitHub

Memcached Server

Amazon Hackathon project focused on product classification using ALBERT models.

Utilized ALBERT models and implemented ensemble technique to enhance accuracy by 10%. Employed synthetic optimized PyTorch datasets for improved performance.

View on GitHub

Big Small Data Analytics

Probabilistic data structures that can process and search very large amount of data super fast, with little loss of accuracy..

A comprehensive data analytics solution that handles both big and small data processing:

  • Implements the following alogorithms
  • MinHash LSH MinHash, Weighted MinHash Jaccard Threshold MinHash LSH Forest MinHash
View on GitHub

Django Food Delivery Platform

Full-stack multi-vendor food delivery platform with advanced features including real-time order tracking, payment integration, and location-based services.

A comprehensive food delivery platform built with Django, featuring:

Core Features
  • Asynchronous shopping basket functionality
  • Integrated Razorpay and PayPal payment gateways
  • PostgreSQL database integration
  • Custom user authentication framework
  • Automated email notifications system
Vendor Management
  • Vendor registration and authentication system
  • Admin review interface for vendor approval
  • Custom vendor control panels
  • Restaurant profile management
  • Menu and food item CRUD operations
User Experience
  • Google Maps integration for address suggestions
  • Location-based restaurant search
  • Advanced search capabilities
  • Streamlined checkout process
  • Real-time order tracking
Technical Highlights
  • Django event triggers implementation
  • Dynamic tax calculation system
  • Custom order management workflow
  • Email template integration
  • Responsive design implementation
View on GitHub

LazaOS - Operating System Development

A custom operating system implementation focusing on fundamental OS concepts and low-level system programming.

An educational operating system project demonstrating core OS concepts:

Kernel Development
  • Bootloader implementation in Assembly
  • Memory management and paging
  • Interrupt handling system
  • Process scheduling
  • System call interface
Device Management
  • Device driver architecture
  • Keyboard input handling
  • Display output management
  • Basic I/O operations
  • Hardware abstraction layer
File System
  • Basic file system implementation
  • File operations (read/write)
  • Directory structure
  • File descriptors
  • Storage management
System Features
  • Multi-tasking support
  • Memory protection
  • User/kernel mode separation
  • Basic shell implementation
  • Error handling system
Technical Implementation
  • Written in C++ and x86 Assembly
  • GRUB bootloader integration
  • Custom build system and toolchain
  • QEMU/Bochs emulator support
Assembly C++ OS Development Kernel Programming System Architecture Low-level Programming
Project Highlights:
  • Successfully implemented basic kernel functionality
  • Developed custom bootloader and memory management
  • Created working file system implementation
  • Implemented basic shell interface
View on GitHub

File Zipper

A sophisticated file compression tool implementing the Huffman coding algorithm for efficient text file compression and decompression. Features automatic file handling and optimized encoding/decoding processes.

Core Features
  • Huffman encoding algorithm
  • Text file compression
  • Automatic file decoding
  • Download management
Technical Details
  • Binary tree implementation
  • Priority queue management
  • File stream handling
  • Optimized compression ratios
HTML5
CSS3
JavaScript
PHP
View Source

Deep Learning Paper Implementations

Collection of implemented research papers in deep learning, focusing on computer vision and neural networks.

A comprehensive repository of deep learning paper implementations, featuring:

Vision Transformers
  • Implemented Vision Transformer (ViT) architecture
  • Attention mechanism for image processing
  • Patch-based image tokenization
  • Positional embeddings implementation
  • Multi-head self-attention modules
Neural Networks
  • Custom loss functions implementation
  • Gradient computation and backpropagation
  • Layer normalization techniques
  • Various activation functions
  • Optimization algorithms
Training & Evaluation
  • Training pipeline development
  • Performance metrics implementation
  • Model evaluation frameworks
  • Hyperparameter tuning
  • Cross-validation strategies
Technical Stack
  • PyTorch for model implementation
  • NumPy for numerical computations
  • Matplotlib for visualizations
  • Jupyter notebooks for experiments
  • Git for version control
PyTorch Computer Vision Transformers Neural Networks Deep Learning
View on GitHub

User Rating Prediction

Advanced rating prediction system with multilingual support.

Developed a model with feature engineering, created RAG pipeline for multilingual queries, utilized FAISS for vector database management, and containerized with Docker.

View on GitHub

Django Recommendation System

Content-based recommendation system built with Django.

Developed a Django website using advanced recommendation algorithms including content-based filtering, FunkSVD, and popularity-based recommenders.

View on GitHub

Customer Churn Prediction

ML model for predicting customer churn with high accuracy.

Developed a predictive model achieving 85% accuracy through advanced feature engineering and machine learning techniques.

View on GitHub

Skills

Deep Learning
Statistics
C++
Docker
Databases
LLD
Django
NodeJs, ExpressJs
AWS
Nginx

Experience

Software Engineer | Deloitte (Full Stack with Backend Heavy)

Deloitte

July 2023 - Present | Hyderabad

Developed full-stack applications using Node.js, JavaScript, created REST APIs, and deployed on AWS Built secure authentication using JWT, configured middleware for logging and error handling Developed webhooks for chatbots using Python SDK, implemented Business Entity Recognition Managed 6 large custom UI applications with focus on security best practices and data integrity

Freelance Machine Learning Developer

Unify.ai

February 2023 - May 2023 | Remote

Implemented functions in TensorFlow, JAX, NumPy, Torch, PaddlePaddle; increased code coverage by 10 • Led unit-testing with Hypothesis, managed array-api-test suite achieving 80 • Wrote Ivy OpConverters for TensorRT backend, guided contributors through large codebase

Publications

SPECTRUM 2020 - Application of Machine Learning in Medicine

Discussed how various deep learning models solve complex real-life problems

Read Publication

IEMATICS 2020 - Homomorphic Encryption

Research on data compression techniques for secure transmission and storage

Read Publication

Blog Posts

Featured

Understanding Transformers: A Deep Dive into Attention Mechanisms

An in-depth exploration of transformer architecture and how attention mechanisms revolutionized natural language processing.

Oct 15, 2023 15 min read Machine Learning, NLP
Featured blog post

Introduction to Transformers

The Transformer architecture, introduced in the landmark paper "Attention Is All You Need," has revolutionized natural language processing and beyond. This blog post provides a comprehensive overview of how Transformers work and why they're so effective.

Key Components

  • Self-Attention Mechanism: Allows the model to weigh the importance of different words in relation to each other
  • Multi-Head Attention: Enables the model to focus on different aspects of the input simultaneously
  • Position Encodings: Provides sequential information without recurrence

How Attention Works

The attention mechanism can be broken down into three main components:

  1. Query vectors: What we're looking for
  2. Key vectors: What we have to match against
  3. Value vectors: The actual content we want to aggregate

Practical Applications

Transformers have found success in various applications:

  • Machine Translation
  • Text Summarization
  • Question Answering
  • Code Generation
Read Full Article
Blog post

Building Scalable Backend Systems with Node.js

Best practices and architectural patterns for building high-performance Node.js applications.

Sep 28, 2023 10 min read

Scalability in Node.js

Building scalable backend systems requires careful consideration of architecture, performance, and maintainability. This post explores proven patterns and best practices.

Key Architectural Patterns

  • Microservices Architecture: Breaking down complex applications
  • Event-Driven Design: Handling asynchronous operations
  • Caching Strategies: Improving response times

Performance Optimization

Learn about crucial performance optimization techniques:

  • Database query optimization
  • Memory management
  • Load balancing
Read Full Article
View All Posts

Achievements

Contact Me

Let's Connect

bishnupadamajumder32@gmail.com
Hyderabad, India

Education

Institute of Engineering and Management

Electronics and Communication Engineering

2019 - 2023

CGPA: 8.94