Ehsan Irshad

Experience

• 

  Virtual COMSATS

  Mar 2015 - Apr 2019

  Research Associate

  👉Supervised and implemented the FYP Project "Stock Market Price Prediction using LSTM based Approaches"-RMSE and MAPE of 5%👉Supervised and Implemented the FYP "Sentiment Analysis Using Twitter Data"👉Supervised and Implemented FYP Project "Crime Prediction and Classification in San Francisco City (May 2017-Aug 2017)" Achieved ~98% (2.6 improvement) accuracy using different classifier like Decision Trees, Random Forest, SVM and AutoMLP. Public ranking at Kaggle was 850 . Principal Component Analysis (PCA) and Single Value Decomposition (SVD) used for data pre-processing 👉Supervised and Implemented the FYP Project " Cabs Demand Prediction of UBER Data Movement API" using sklearn, and pandas. Achieved RMSE or MAPE 6 %. Show less

• 

  PIEAS University and Interactive Group

  Apr 2019 - Oct 2020

  Big Data Analyst (Technical Team Lead)

  👉Prepared Raw GPS trajectories from Vehicle Tracking company and make them associated with Roads and error is MINIMIZED👉Urban planning: Calculated Attractiveness of Polygon is MEASURED by taking the causality relationship between linestring and ploygonOSRM routing server is setup for getting nearest nodes and ways information for Assigning the ground truth to Raw GPS Trajectories👉Traffic Congestion Prediction and VISUALIZATION Platform (HEC- Technology DEVELOPMENT Fund) Vehicle Speed Prediction using Raw GPS Trajectories : RMSE, MAPE 12% 👉 Developed the Data Integration Pipeline using Docker and Kubernetes 👉Volume Consistency in Docker was done in Postgresql Show less

• 

  Government of Punjab

  Oct 2020 - Sept 2021

  IT Manager Robotics

  👉Air quality data (pm2.5, PM10, tvoc etc.) acquisition from sensors using microcontrollers and raspberry pi on board computers for Lahore Punjab👉Data Acquisition from Google Earth Engine (GEE) in Tiff files of air Quality on Time series based - 195 location of Lahore👉Air Quality Data Preparation from Tiff to CSV (Made YouTube Tutorial as well)Smog Prediction in Lahore using the above prepared data - RMSE is 2%👉Building Detection from Satellite imageries for predicting the change in growth of urban areas using UNET Model - Spacenet Challenge. 👉Orange Disease Detection and Recognition using Mobilenet Architecture👉Developed real-time data dashboard using ELK Stack👉Developed Air Quality Data Dashboard using Django Plotly dash👉Time series Air Quality Forecasting using ARIMA, Seasonal ARIMA -MAPE 3%import (procurement) of drones components for data acquisition of SAR (Synthetic Aperture Radar) imageries in Punjab (Pakistan)👉market and specification analysis of drones technologies for procurement👉For Building Detection i used Solaris SATELLITE images pre-processing libraryEO-LEARN SATELLITE images deep learning library for detection and recognition of building in satellite imageries Show less

• 

  National Language Processing Laboratory Islamabad

  Sept 2021 - now

  Program Manager OCR

  👉Developed URDU OCR End to End Pipeline Using Tesseract and LSTM - WER 20%👉Developed URDU Data Scrapping Pipeline and Data Augmentation using 512 different Fonts👉Developed Urdu Speech Recognition App using Transformer based approaches in Huggingface -9 speakers 👉Speech Data Augmentation using Librosa and Essentia for Noise removal and Noise AdditionSpeech Cloning for Speech data Augmentation Machine Translation using Transformer based approaches like seq2seq modelling - WER and Bleu 15%👉Docker Speech Data Preparation Pipeline Developed Flask API for URDU OCRWord Segmentation of URDU Ligatures👉Developed Real-time Data acquisition pipeline for machine translation using Kafka Using Fuzzy Based approached like levenshtein distance n-gram and cosine similarity for machine translation data preparation. 👉Deployed Machine Learning Models and Developed Sophisticated system design architecture to support million of users using Load balancing consistent hashing and consensus algorithms. 👉Organization of Team Task by maintaining the product backlog- Running Sprints for Data Science Activities Show less