Skip to content
View geetu040's full-sized avatar

Highlights

  • Pro

Block or report geetu040

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.

Please don't include any personal information such as legal names or email addresses. Maximum 100 characters, markdown supported. This note will be visible to only you.
Report abuse

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

Report abuse
geetu040/README.md

👤 About Me

Github Stats

Hi 👋

I am Armaghan,
Software Engineer from NUST,
interested in AI and Machine Learning.

I work mostly in the space of
Computer Vision,
Large Language Models,
and AI-Agents.

And sometimes, you'll find me
contributing to open source as well.


🛠️ My daily development tools


🚀 Career Highlights

See all experiences at Linkedin

Machine Learning Engineer @ ecoSPECS

  • Working under European Summer of Code (ESoC) program.
  • Automating document generation and validation in biotech/pharma domain.
  • And accelerating qualification processes for documents using AI agents and LLMs.

Google Summer of Code & Core developer @ Sktime

  • Integrated Sktime with Deep Learning Backends (Pytorch, Huggingface, PEFT)
  • Added Pre-trained Foundation Models
  • Created 20+ Pull Requests and Reviewed 15+ Pull Requests

DAAD Research Intern @ Hochschule RheinMain

  • Selected for Research Internship under DAAD at Hochschule RheinMain
  • Researched on Knowledge Distillation Techniques for Document Understanding
  • Contributed 2 Research Publications

Research Intern @ TUKL-NUST

  • Implemented Convolution and Recurrent Neural Networks for EEG Pathology Detection.
  • Researched AI-based fire detection and prevention
  • Led data analysis for an e-commerce platform (Markaz App)
  • Developed satellite-based tree count system for Margalla Hills (Govt. project)

🔥 Projects

SlimDoc: Lightweight Distillation of Document Transformer Models

  • This paper introduces SlimDoc, a parameter-efficient distillation method for multimodal document transformers that achieves strong performance without labeled data, outperforming standard fine-tuning and single-stage distillation across multiple document understanding tasks.
  • Code: marcel-lamott/SlimDoc

Leveraging Distillation Techniques for Document Understanding

  • This paper explores distilling document understanding capabilities from ChatGPT into FLAN-T5 using labeling and curriculum learning to create a scalable, efficient alternative to resource-heavy LLMs.
  • Paper Link: arXiv:2409.11282

Google Summer of Code @ Sktime, Code Submission

DepthPro: Beyond Depth Estimation

  • Depth Estimation model, DepthPro by Apple, experimented for Image Segmentation and Image Super Resolution.
  • Project Link: geetu040/depthpro-beyond-depth
More Projects
Project Description
Pixa AKS Pixa is our cloud computing project. A gallery web app that involves orchestrating communication among various microservices, deploying them onto Github Container Registry and Azure Kubernetes Service, and ultimately automating the deployment workflow using Github Actions.
Cloud Computing - Azure - Kubernetes
Article on the Project Architechture - Project Code
Pymodels Series of Data Science projects, where I have created Computer Vision, NLP models and Recommender Systems from scratch
Data Science - Pytorch - Tensorflow - Scikit-Learn
Project Demo - Project Code - Project APIs on Huggingface - Amazon Reviews - Movie Reviews
Visual DSA A Visual Implementation of Data Structures and Algorithms including Trees, Graph and Sorting Algorithms created in React.
DSA - React.js
Project Demo - Project Code
FM Connect.AI Enables document uploads and chatbot interaction. It integrates multiple databases, APIs, and is hosted on Google Cloud Platform, using technologies like Flask, Google Cloud, and Pinecone.
Cloud Computing - Aritificial Intelligence - GCP
More on Fiverr
NUST Bank Chatbot LLM-based customer support chatbot for a fictional bank, built as part of our LLM project at NUST. Uses RAG, small open-source models, and supports real-time document updates and local inference.
LLMs - RAG - Chatbot
Project Code
Book Recommendation - Graph Embeddings - DeepWalk Using Graph Embeddings and DeepWalk algorithm to find similar books applying collaborative filtering
Data Science - DSA
Code
Covid Data Analysis I visualized and analyzed different graphs on Covid Data using advanced libraries for EDA in python
Exploratory Data Analysis
Code 1 - Code 2 - Code 3 - EDA Ecommerce Consumer Behaviour
Intelli Snake The snakes uses Convolution Network based custom trained Deep Learning model to avoid obstacles (walls, blocks and its body) to find the shortest path to food.
Data Science - Tensorflow
Project Code - Project Demo - Huggingface Model
Audio Transcribing and Diarize Using whisper and pyannote to transcribe and diarize youtube videos
Aritificial Intelligence
Code - More on Fiver
UniFinder A college finder web app using Cloud Computing and Continuous Integration. Automating Deployment of Python, PHP, Node js and React app on Azure App Services and Functions.
Cloud Computing - Azure
Project Code - Article on Azure MySQL
Data Structures and Algorithms DSA implementation in Python, C++, Java and Javascript
DSA
Code
Text2Video Using state of art models for text to video generation, developing custom API and hosting on Paperspace GPU
Cloud Computing - Artificial Intelligence
More on Fiverr
Pricegram E-commerce products EDA, clustering, recommender and search engine
Data Science
Dataset - Search Engine - Search Engine using Qdrant - Search Engine Interface
Portfolio Website Website developed for portfolio projects
Web Development - Next.js
Code - Demo
Forest Fire Mitigation As a part of my lab project, I researched on the methods and state of art techniques in the domain of mitigating forest fires
Research
Report
EEG Pathology Detection Applied the techniques of Convolution and Recurrent Networks in implementing research based solutions for pathology detection in EEG time-series data
Data Science - Pytorch
Reference Code
Quora Search Engine Using Qdrant Example of creating a search engine in no time Artificial Intelligence
Code - Article
Assembly Code Simple code pieces during my practise for coding in assembly
Assembly
Code
News App Uses an API to get news of different categories and shows in a descent User Interface.
Web Development - React.js
Project Code - Project Demo
Text Utils A simple utility for text where you can capitalize, count etc.
Web Development - React.js
Project Code - Project Demo
Todo App A simple todo application
Web Development - React.js
Project Code - Project Demo
Static HTML Pages I have hosted my simple apps based on static web pages that use only HTML, CSS, JavaScript, Bootstrap, JQuery or built from React and Next js projects. These include simple games and clones.
Web Development
Code - Demo
Find Closest City Laravel implementation of a website that uses map service to find cities closest to the selection
Web Development - Laravel
Code
Web Crawler PHP implementation of a basic web crawler
Web Development - PHP
Code
Deep Learning Courses My practice code while completing the courses on coursera
Data Science
Code
Pacman Pacman game implemented using tkinter in python
Tkinter
Code

🌍 Open Source Contributions

I mostly contribute state-of-the-art models to popular open-source libraries, for my understanding, and better accessibility and usability for others.

Models I Implemented

  • [ huggingface/transformers ] MiniMax-Text-01 by MiniMaxAI ( PR . Docs . Paper )
  • [ huggingface/transformers ] DepthPro by Apple ( PR . Docs . Paper )
  • [ sktime/sktime ] TinyTimeMixer by Granite ( PR . Docs . Paper )
  • [ sktime/sktime ] TimesFM by Google ( PR . Docs . Paper )
  • [ sktime/sktime ] MVTSTransformerClassifier by IBM ( PR . Docs . Paper )
  • [ sktime/sktime ] LTSFTransformerForecaster by CureLab ( PR . Docs . Paper )

Models I Helped Implement

  • [ sktime/sktime ] TinyTimeMixer by Granite ( PR . Docs . Paper )
  • [ sktime/sktime ] SciNet by Granite ( PR . Docs . Paper )
  • [ sktime/sktime ] ES-RNN by Granite ( PR . Docs . Paper )

Other Fixes & Enhancements

  • [ huggingface/transformers ] Expose SamVisionModel ( PR )
  • [ huggingface/transformers ] Fix/Refactor Sam ( PR )
  • [ sktime/sktime ] PEFT methods in HFTransformersForecaster ( PR )
  • [ qdrant/qdrant-client ] GRPC Channel-Level Compression ( PR )
  • [ qdrant/fastembed ] Support MiniLM-L12-v2 ( PR )

Find all my authored PRs and issues here: ( huggingface/transformers . sktime/sktime )


📫 Lets connect

Pinned Loading

  1. sktime/sktime sktime/sktime Public

    A unified framework for machine learning with time series

    Python 9.1k 1.6k

  2. Google-Summer-of-Code-2024-Sktime Google-Summer-of-Code-2024-Sktime Public

    Final Submission for Google Summer of Code, 2024 @sktime

    Jupyter Notebook 1 1

  3. depthpro-beyond-depth depthpro-beyond-depth Public

    Depth Estimation model, DepthPro by Apple, trained for Image Segmentation and Image Super Resolution.

    Jupyter Notebook 10 1

  4. visual-dsa visual-dsa Public

    A Visual Implementation of Data Structures and Algorithms

    JavaScript

  5. pymodels pymodels Public

    Series of Data Science projects, where I have created Computer Vision, NLP models and Recommender Systems from scratch

    Jupyter Notebook

  6. pixa pixa Public

    Pixa is our cloud computing project. A gallery web app that involves orchestrating communication among various microservices, deploying them onto Github Container Registry and Azure Kubernetes Serv…

    JavaScript 1