Skip to content
View anushkadas-coder's full-sized avatar
💭
overly caffeinated ☕︎
💭
overly caffeinated ☕︎
2 followers · 1 following

Achievements

Achievement: Pair Extraordinairex4Achievement: Pull Sharkx3Achievement: QuickdrawAchievement: YOLO

Achievements

Achievement: Pair Extraordinairex4Achievement: Pull Sharkx3Achievement: QuickdrawAchievement: YOLO

Block or report anushkadas-coder

Block user

Prevent this user from interacting with your repositories and sending you notifications. Learn more about blocking users.

You must be logged in to block users.

Maximum 250 characters. Please don’t include any personal information such as legal names or email addresses. Markdown is supported. This note will only be visible to you.
Report abuse

Contact GitHub support about this user’s behavior. Learn more about reporting abuse.

Report abuse
anushkadas-coder/README.md

Coding character

Typing SVG

Gmail LinkedIn Google Developers HackerRank

I'm an IT undergrad student from Kolkata, India and an active membe of Google's Women Techmakers and IEEE Robotics and Automation Technical Committee on Machine Learning for Automation. I'm an aspiring ML/DL Engineer with a strong focus on Artificial Intelligence and Machine Learning, I build intelligent systems and full-stack AI applications to solve real-world challenges. Alongside my passion for AIML, I have hands-on experience in Web Development and enjoy problem-solving through Data Structures and Algorithms.


Currently seeking opportunities in Software Engineering and Machine Learning, and additionally in Data Analysis, where I can build impactful, data-driven solutions.

My GitHub Stats

Research & Publications

Architectural Paradigms for Sovereign AI: Mitigating API Fragility through Quantized Low-Rank Adaptation and 1.58-bit Ternary Models

Read Paper | GitHub Repo
Domain: QLoRA, BitNet 1.58-bit, Sovereign AI, PEFT
Investigates the synergy between 4-bit NormalFloat (NF4) quantization, native 1.58-bit ternary architectures, and Low-Rank Adaptation (LoRA) to deploy Sovereign AI via SLMs. Achieved a 93% reduction in inference OpEx for 7B–14B parameter models with a TCO break-even point at 450 million tokens processed compared to proprietary APIs.

Isolating Synthetic Fingerprints: A Frequency-Domain Approach to Robust Deepfake Detection using 2D FFT and Lightweight CNNs

Read Paper | GitHub Repo
Domain: 2D FFT, ResNet-18, PyTorch, Deepfake Detection
A lightweight deepfake detection engine utilizing a micro-blur preprocessing pipeline and 2D Fast Fourier Transform (FFT) to extract underlying spectral magnitude maps. Outperforms spatial baselines, achieving 99.12% accuracy on FaceForensics++ while maintaining real-time inference at 45 FPS on edge devices.

My Skills

Languages

Python Java C++ SQL JavaScript

Libraries & Frameworks

Scikit-learn Pandas NumPy XGBoost Flask React OpenAI HTML5 CSS3

Tools & Platforms

AWS Streamlit Git Jupyter Power BI Windows VS Code

Experience

Period Role Company Highlights
Jan 2025 - Present Member of Google's Women Techmakers Google For Developers Engage in global community supporting women in tech through events, learning resources, and networking.
Dec 2025 - Present Google Cloud Innovator Google For Developers Stay connected with Google Cloud updates, expert sessions, and community insights to strengthen cloud skills.
Jan 2026 (1 month) Blockchain Risk Developer Zetheta Algorithms Private Limited Engineered a vulnerability scanner for Solidity contracts to detect critical flaws like Reentrancy and Access Control.
July 2025 - Sept 2025 Open-Source Contributor GSSoc '25 Resolved bugs and integrated new feature modules across 4 open-source repositories via GitHub.
Aug 2025 (1 month) Web Development Intern CodSoft Developed and enhanced 6+ responsive web pages using HTML, CSS, and JavaScript, significantly improving UI accessibility.

Coder

Top Projects

Federated Fresh: Hybrid AI Orchestration Terminal

Live Demo | GitHub Repo
Technologies: FastAPI, Google Gemini 2.5 Flash, ChromaDB, Python, DuckDuckGo API, CSS3/JS
High-performance, multi-modal AI Terminal featuring Local RAG (ChromaDB), real-time Web Search integration, and Google Gemini 2.5 Flash reasoning. Optimized for low-memory (512MB RAM) cloud deployment.

DeepfakeDetector: Frequency-Based AI Inference Engine

Live Demo | GitHub Repo
Technologies: React, FastAPI, Python, PyTorch, OpenCV, Docker, 2D FFT, ResNet-18, Tailwind CSS
Neural Authenticity Engine: A full-stack deep learning pipeline identifying synthetic media via spectral fingerprinting and 2D FFT. Built with PyTorch (ResNet-18), featuring a Gaussian-denoised inference engine and temporal aggregation.

Deweathering Engine

Live Demo | GitHub Repo
Technologies: Python, React, FastAPI, OpenCV
A document restoration engine leveraging the Inexact Augmented Lagrange Multiplier (IALM) algorithm and Robust PCA (RPCA) to separate sparse text matrices from low-rank background noise in heavily degraded documents.

Deep RL Double Q-Networks for Atari

GitHub Repo
Technologies: Python, Deep Reinforcement Learning, Convolutional Neural Networks (CNN)
A reinforcement learning agent utilizing a Double Deep Q-Network (DDQN) architecture to autonomously master Atari Space Invaders directly from raw pixel input, optimized with Experience Replay.

Customer Churn Prediction

Live Demo | GitHub Repo
Technologies: Python, Scikit-learn, XGBoost, Streamlit
Developed an end-to-end machine learning model to predict telecom customer retention, achieving a 67% recall rate and an AUC score of 0.82. Engine deployed as an interactive web application on Streamlit Community Cloud.

Honors, Awards & Certifications

View Certifications (Click to Expand)
  • Data Analytics Job Simulation - Deloitte Australia
  • Software Engineering Job Simulation - JPMorgan Chase & Co.
  • Introduction to agent skills - Anthropic Academy
  • Train and manage a machine learning model with Azure Machine Learning - Microsoft
  • Understand data science for Machine Learning - Microsoft
  • Python for Data Science - IBM
  • GenAI Powered Data Analytics Job Simulation - Tata
View Honors & Awards (Click to Expand)
  • HackerRank: Gold Level (5 Star) in Python, Silver Level (3 Star) in Java.
  • SIH '25 Semi-Finalist: Authenticity Validator for Academia: 'Digi-Pramaan', Top Finalist out of 250+ competing teams in the college internal round for the Smart India Hackathon(SIH) 2025.
  • Problem Solving: Solved 350+ problems on LeetCode and HackerRank, improving algorithmic thinking and coding efficiency.
  • Extracurricular: Active member of Samaritans social club since first year, contributing to community initiatives and collaborative events and occasionally participate in debate and quiz competitions, showcasing strong communication and critical thinking skills.

Pinned Loading

  1. Customer-Churn-Prediction Customer-Churn-Prediction Public

    An end-to-end project to predict customer churn using Python and XGBoost. Identifies key attrition drivers through data analysis and predictive modeling to provide actionable business insights.

    Python 2

  2. deweathering-engine deweathering-engine Public

    A document restoration engine using Robust PCA (RPCA) to separate sparse text matrices from low-rank background noise. Built with FastAPI and React.

    JavaScript 2

  3. deepfake-detection-engine deepfake-detection-engine Public

    Neural Authenticity Engine: A full-stack deep learning pipeline identifying synthetic media via spectral fingerprinting and 2D FFT. Built with PyTorch (ResNet-18), it features a Gaussian-denoised i…

    Python 2

  4. Federated_Fresh Federated_Fresh Public

    A high-performance, multi-modal AI Terminal featuring Local RAG (ChromaDB), real-time Web Search integration, and Google Gemini 2.5 Flash reasoning. Optimized for low-memory (512MB RAM) cloud deplo…

    Python 2

  5. DCO6_synthesizer DCO6_synthesizer Public

    A 6-voice polyphonic VST3 synthesizer built from scratch in C++ and the JUCE framework, featuring a 4-pole ladder filter, BBD stereo chorus, and real-time FFT analysis.

    C++ 2

  6. Deep-Reinforcement-Learning-with-Double-Q-Networks-on-Atari-Games Deep-Reinforcement-Learning-with-Double-Q-Networks-on-Atari-Games Public

    This project uses deep RL to train an agent that can play Atari game named Space Invaders.

    Jupyter Notebook 2