Projects

Research and software

Research and software projects, grouped by theme.

104 projects

Current research

6 projects

QDSpace web app

Web platform for QuantumDotSpace: a nanocrystal builder with facet, shape and ligand-passivation control, and a library of DFT-relaxed perovskite, II–VI and III–V structures with interactive PDOS and COOP visualisation.

  • Svelte
  • MatterViz

github.com/nlesc-nano/QDSpaceWebApp

QDSpace Desktop

Standalone desktop version of the platform, built with Tauri and Rust, with a built-in local property calculator based on MACE.

  • MACE
  • Tauri
  • Rust
  • Desktop

github.com/nlesc-nano/QDSpaceDesktop

Orchestr.AI

Training, inference, and benchmarking across SchNet, PaiNN, SO3net, FieldSchNet, NequIP, Allegro, and MACE.

One YAML file, SLURM, and multi-GPU. Release v0.2.0 on Zenodo.
  • SchNet
  • PaiNN
  • NequIP
  • Allegro
  • MACE

github.com/nlesc-nano/Orchestr.AI

Quantum-dot force fields

Developing machine learning force fields (MLFF) for quantum dots to enable efficient atomistic simulations of nanomaterials.

Data, training, and evaluation pipelines on Python and HPC environments.
  • Python
  • HPC

Foundation potential for quantum dots

Machine-learning interatomic potential foundation model for quantum dots: one potential rather than a new force field for each system.

Drug discovery

Solubility, bioactivity, and pIC50 prediction from molecular structure.

Individual studies are listed under Cheminformatics.
  • RDKit
  • ChemBERTa

Multi-agent

3 projects

Automated research agents

Multi-agent LLM system for automated research using Gemini 1.5 Pro and Google search tools.

  • Gemini 1.5 Pro

CrewAI market research

CrewAI-based multi-agent system using Mistral-7B for market research and content strategy.

  • CrewAI
  • Mistral-7B

LangGraph research assistant

Multi-agent research assistant using LangGraph for automated web research and content synthesis.

  • LangGraph

Generative AI

5 projects

Synthetic dataset generation

Synthetic dataset generation using Generative Adversarial Networks (GANs) for privacy-preserving machine learning.

  • GAN

LoRA GPT-2 fine-tuning

LoRA-based fine-tuning of GPT-2 for positive movie review generation using the IMDB dataset.

  • LoRA
  • GPT-2
  • IMDB

LLM code generation

LLM-based code generation system using the Salesforce CodeGen model, Hugging Face Transformers, and GitHub API.

  • CodeGen
  • Hugging Face Transformers
  • GitHub API

Transformer fine-tuning

Fine-tuning Transformer models including BERT for text classification and GPT-2 for text generation.

  • BERT
  • GPT-2

Text generation and summarization

Text generation and summarization using GPT-2, GPT-Neo-2.7B, Pegasus, BART and T5.

  • GPT-2
  • GPT-Neo-2.7B
  • Pegasus
  • BART
  • T5

Document intelligence

2 projects

LLM document analysis platform

LLM-powered document analysis platform for text extraction, summarization, and insight generation.

T5 and RoBERTa analysis

Document analysis system using T5 and RoBERTa models with Hugging Face Transformers.

  • T5
  • RoBERTa
  • Hugging Face Transformers

RAG

4 projects

FAISS RAG system

Retrieval-augmented generation (RAG) system using FLAN-T5, Sentence Transformers, FAISS, and LangChain.

  • FLAN-T5
  • Sentence Transformers
  • FAISS
  • LangChain

Real-time web search assistant

Real-time web search AI assistant using Llama-3, LangChain, and DuckDuckGo search.

  • Llama-3
  • LangChain
  • DuckDuckGo

Hybrid RAG expert assistant

Hybrid RAG-based expert assistant using LangChain, ChromaDB, and BM25 retrieval with Groq LLM inference and a Chainlit interactive interface.

  • LangChain
  • ChromaDB
  • BM25
  • Groq
  • Chainlit

OpenAI retrieval QA

Retrieval-Augmented Question Answering using LangChain, OpenAI and ChromaDB.

  • LangChain
  • OpenAI
  • ChromaDB

Applied LLM

1 project

Resume screening system

Intelligent resume screening and skill gap analysis system using Llama-3, Ollama, and PyMuPDF.

  • Llama-3
  • Ollama
  • PyMuPDF

Advanced NLP

4 projects

Sentiment and emotion modeling

Sentiment analysis and emotion modeling using BERT, EmoRoBERTa and Hugging Face Transformers.

  • BERT
  • EmoRoBERTa
  • Hugging Face Transformers

SQuAD question generation

Question generation and question answering systems using Transformer models for the SQuAD dataset and Hugging Face frameworks, including multitask QA generation.

  • SQuAD
  • Hugging Face Transformers

LLM NLP Streamlit tools

LLM applications and NLP tools covering summarization, question generation and answering, answer evaluation, YouTube transcript summarization, and document classification.

  • Text summarization application using text-davinci-003 with Streamlit
  • Question generation and answering applications using text-davinci-003, OpenAI and LangChain with Streamlit
  • Automated question-answer evaluation using Streamlit
  • YouTube transcript summarization using LangChain
  • Document classification using BERT
  • text-davinci-003
  • OpenAI
  • LangChain
  • Streamlit
  • BERT

Semantic text similarity

Semantic text analysis and similarity modeling using Sentence Transformers (all-MiniLM-L6-v2, all-MiniLM-L12-v2, all-distilroberta-v1, all-mpnet-base-v2) and SpaCy.

  • Sentence Transformers
  • spaCy

Cheminformatics

12 projects

Molecular solubility prediction

Molecular solubility prediction using Linear Regression and LSTM models.

  • Linear Regression
  • LSTM

PRC2 inhibitor prediction

PRC2 inhibitor prediction using PaDEL molecular descriptors, feature engineering, machine learning, virtual screening, molecular docking and molecular dynamics.

  • PaDEL
  • Virtual screening
  • Docking
  • Molecular dynamics

MPNN ChemBERTa pIC50

Molecular pIC50 prediction using a hybrid Message Passing Neural Network (MPNN) with pretrained ChemBERTa fusion and a Streamlit QSAR dashboard.

  • MPNN
  • ChemBERTa
  • Streamlit

Multi-task molecular activity

Multi-task deep learning framework for multi-target molecular activity prediction using Mol2Vec embeddings with shared representation and task-specific heads for simultaneous activity classification and pIC50 regression, incorporating focal loss and large-scale virtual screening.

  • Mol2Vec
  • Focal loss
  • Virtual screening

End-to-end QSAR platform

Interactive end-to-end QSAR platform for molecular data curation, feature engineering, machine learning modeling, and virtual screening using RDKit descriptors, molecular fingerprints, and AutoML (PyCaret).

  • RDKit
  • PyCaret
  • AutoML

SMILES transformer QSAR

Transformer-based molecular pIC50 prediction from SMILES using fine-tuned BERT and pretrained ChemBERTa models for sequence-based QSAR modeling.

  • BERT
  • ChemBERTa
  • SMILES

LSTM SMILES embeddings

SMILES representation learning using LSTM embeddings combined with Random Forest for molecular bioactivity classification.

  • LSTM
  • Random Forest
  • SMILES

ChemBERTa and ChemGPT

Transformer-based molecular representation learning using ChemBERTa and ChemGPT embeddings with Random Forest for molecular bioactivity classification using pretrained models (ChemGPT-4.7M, ChemGPT-1.2B, ChemBERTa-zinc-base-v1).

  • ChemBERTa
  • ChemGPT
  • Random Forest

ChemBERTa RNN bioactivity

SMILES-based molecular bioactivity classification using ChemBERTa embeddings with recurrent neural networks (RNN).

  • ChemBERTa
  • RNN
  • SMILES

Stacked ensemble meta-learning

Stacked ensemble meta-learning model for molecular bioactivity classification using Morgan fingerprints.

  • Morgan fingerprints

Graph attention pIC50

Molecular pIC50 prediction using graph attention networks (GAT) with molecular graph representations.

  • GAT

Chemprop bond MPNN

Molecular bioactivity classification using bond message passing neural networks with Chemprop.

  • Chemprop

Multilingual NLP

10 projects

Advanced Naive Bayes sentiment

Implementation of the paper "Advanced Naïve Bayes Algorithm Design with Part-of-Speech Tagger on Sentiment Analysis".

  • Naive Bayes
  • POS tagger

Roman Urdu sentiment analysis

Deep learning-based sentiment analysis for Roman Urdu text.

Emotion detection from text

Emotion detection from text using Random Forest and LSTM.

  • Random Forest
  • LSTM

Roman Urdu lexical normalization

Lexical normalization of Roman Urdu text.

Twitter PSL sentiment analysis

Sentiment analysis on Twitter data (PSL) for region-specific keyword analysis.

  • Twitter
  • PSL

Named entity recognition

Named Entity Recognition using LSTM.

  • LSTM

IMDb sentiment analysis

Sentiment analysis on the IMDb movie reviews dataset.

  • IMDb

Insincerity detection

Insincerity detection in textual data using LSTM and GRU.

  • LSTM
  • GRU

Toxic comment classification

Toxic comment classification using Logistic Regression, SVM and Random Forest.

  • Logistic Regression
  • SVM
  • Random Forest

Part-of-speech tagging

Part-of-Speech tagging using Decision Tree, Hidden Markov Model, Artificial Neural Networks and LSTM.

  • Decision Tree
  • HMM
  • ANN
  • LSTM

Applied ML & NLP

8 projects

Telecom churn prediction

Telecom customer churn prediction using SVM.

  • SVM

Cardiovascular disease prediction

Cardiovascular disease prediction using AdaBoost and Random Forest.

  • AdaBoost
  • Random Forest

OLX item price prediction

Item price prediction for OLX data using Deep Learning.

  • Deep Learning
  • OLX

Ozone behavior prediction

Ozone (O₃) behavior prediction using PyCaret and AutoML regression models.

  • PyCaret
  • AutoML

Weather type classification

Multi-class weather type classification using PyCaret for improved forecasting.

  • PyCaret

Web phishing detection

Web phishing detection with PyCaret and hyperparameter tuning.

  • PyCaret

Dark pattern detection

Dark pattern detection in e-commerce websites using NLP and machine learning models.

  • NLP
  • Machine learning

Runner injury prediction

Injury prediction for competitive runners using AdaBoost, SVM and Random Forest.

  • AdaBoost
  • SVM
  • Random Forest

Clustering

5 projects

COVID abstract clustering

Document clustering and topic discovery using k-means and Hierarchical Clustering on COVID-19 research abstracts with intra-cluster and inter-cluster evaluation.

  • k-means
  • Hierarchical Clustering

User interest clustering

User interest clustering using k-means on Facebook pages interaction data.

  • k-means

Semantic document clustering

Semantic document clustering system for grouping textual documents based on shared semantic characteristics.

Comment behavioral segmentation

User comment clustering and behavioral segmentation using k-means and text similarity analysis.

  • k-means

EM probabilistic clustering

Probabilistic document clustering using the Expectation–Maximization (EM) algorithm.

  • EM

Keyword extraction

4 projects

Amazon description keywords

Keyword extraction from Amazon product description using YAKE, RAKE-NLTK, TextRank, KeyBERT, SpaCy, RAKE, PKE and MultiRake.

  • YAKE
  • RAKE-NLTK
  • TextRank
  • KeyBERT
  • spaCy
  • PKE
  • MultiRake

Amazon review keywords

Examining various keyword extraction methods for Amazon reviews.

SEO keyword extraction

NLP-based Keyword Extraction for SEO Optimization using Gensim with a web-based Streamlit application.

  • Gensim
  • Streamlit

COVID news keywords

Automated Keyword Extraction for COVID-19 News Articles using KeyBERT and TextRank.

  • KeyBERT
  • TextRank

Topic modeling

4 projects

COVID research topics

Topic modeling and research trend discovery in COVID-19 scientific abstracts using Latent Dirichlet Allocation (LDA) with coherence score tuning.

  • LDA

Amazon review topics

Topic modeling and visualization on Amazon product reviews using LDA and Latent Semantic Analysis (LSA) with t-SNE and pyLDAVis.

  • LDA
  • LSA
  • t-SNE
  • pyLDAvis

Author style topics

Author style topic discovery on the Spooky Author Identification dataset using LDA and Non-negative Matrix Factorization (NMF).

  • LDA
  • NMF

Review topic visualization

Topic distribution analysis and visualization for Amazon product reviews including topic scores and word cloud generation.

  • Word cloud

Data mining

3 projects

Retail recommendation system

Recommendation system for retail transaction data to identify frequent itemsets and suggest optimal product ordering based on sales ratio, seasonal trends and regional demand.

Market basket analysis

Market basket analysis using the Apriori algorithm for frequent itemset mining and association rule discovery in transactional data.

  • Apriori

RapidMiner data mining

Implementation of multiple data mining techniques using RapidMiner.

  • RapidMiner

Computer vision

4 projects

Online cheating detection

Online cheating detection system using ResNet-50 with transfer learning for automated exam monitoring at GIK Institute Pakistan.

  • ResNet-50

Vegetable image classification

Vegetable image classification using Convolutional Neural Networks (CNN).

  • CNN

Pneumonia X-ray detection

Pneumonia detection from chest X-ray images using CNN with the fastai deep learning framework.

  • CNN
  • fastai

VGG16 transfer learning

Transfer learning-based image classification using VGG16 pretrained weights.

  • VGG16

Core ML implementations

7 projects

Enhanced ID3 algorithm

Implementation of the research paper "Enhanced ID3 Algorithm based on Attribute Weighting", written by hand and visualised and animated with Graphviz, Turtle and Anytree. Results are compared against Scikit-learn's decision tree.

  • Graphviz
  • Turtle
  • Anytree
  • Scikit-learn (comparison)

Hidden Markov Model

Implementation of Hidden Markov Model (HMM) for predicting assignment type sequences.

  • HMM

Apriori from scratch

Implementation of the Apriori Algorithm for frequent itemset mining on the Zoo dataset.

  • Apriori
  • Zoo dataset

Naive Bayes from scratch

Implementation of Naïve Bayes classifier.

Genetic algorithm

Implementation of Genetic Algorithm (GA) for team performance optimization.

  • Genetic Algorithm

Regression from scratch

Implementation of Linear Regression and Logistic Regression algorithms from scratch.

Feedforward network from scratch

Implementation of a feedforward neural network from scratch.

Geospatial

4 projects

COVID-19 geospatial analysis

Geospatial analysis of COVID-19 data using GeoPandas and Folium with marker clustering for spatial visualization.

  • GeoPandas
  • Folium

Phone price AutoML maps

Mobile phone price prediction using AutoML with geospatial market visualization using Folium.

  • AutoML
  • Folium

Seller location mapping

Seller location mapping and market analysis using Folium for OLX marketplace data.

  • Folium
  • OLX

Regional market trends

Geospatial visualization of regional market trends using location data extracted from online marketplaces.

  • Folium

Bioinformatics

4 projects

Brain tumor classification

Brain tumor classification using Convolutional Neural Networks (CNN).

  • CNN

Sequence alignment prediction

Sequence alignment prediction using a Fully Convolutional Neural Network (FCN).

  • FCN

Protein family classification

Protein family classification using Random Forest and Logistic Regression.

  • Random Forest
  • Logistic Regression

Sequence classification

Protein, DNA and RNA sequence classification using AdaBoost, Random Forest and deep learning models.

  • AdaBoost
  • Random Forest

Web scraping and automation

14 projects

Social media API scraping

Scraping from different sites using their APIs (Facebook, Twitter, Instagram, YouTube).

  • APIs

Shopping website scrapers

Build Scraper for shopping websites (Amazon, OLX, Alibaba, eBay, AliExpress, Vegas, ishopping, whatmobile.com, just4girls, arysahulatbaza.pk, pakwheels.com).

Poetry website scrapers

Build Scraping for poetry websites (hamariweb, ShayariUrdu, smsmessages101, poetryinurdu.pk, dreamysms, and related Urdu poetry and SMS sites).

Manga chapter automation

Image Scraping and full automation for manganelo.com and mangakakalot.com.

Take a specific title, download all chapter images into chapter folders, and automatically generate HTML for WordPress posting on three websites. The code reduced about three months of work for five people to a single-day run.
  • WordPress

Video site publishing pipeline

Scraping and Automation (automate downloading to posting contents including HTML preparation) for vidcloud9.com and vidstreaming.io.

  • WordPress

OLX brand and price pipeline

OLX pipeline: for a specific item, collect all brands and attributes, resolve brand from title or description, store by brand folder, predict item price, and plot seller locations on a map using Folium.

  • OLX
  • Folium

Facebook group posting

Automate the process of posting contents in multiple Facebook groups with a scheduler.

MS Teams class attendance

Automate the process of class attending (MS Team).

Windows task automation

Automate many window tasks using PyAutoGUI.

  • PyAutoGUI

Form filling automation

Form filling automation with Selenium.

  • Selenium

WordPress posting automation

Automate the process of WordPress posting.

  • WordPress

LinkedIn job market scraping

LinkedIn data scraping for job market and industry trend analysis.

  • LinkedIn

CoinMarketCap data collection

Cryptocurrency market data collection from CoinMarketCap for price and trend monitoring.

  • CoinMarketCap

OLX phone price scraping

Mobile phone price scraping from OLX marketplace.

  • OLX
↑