Hi!
I am Usman Yaqoob
And I Love
Machine Learning / Deep Learning,
Data Science,
Artificial Intelligence :)

Contact

About Me

WHO AM I?

As a Computer Engineering graduate, my passion lies in the realms of Artificial Intelligence, Machine Learning, and Data Science. I derive immense satisfaction from meticulously cleansing datasets, ensuring their optimal utilization. My expertise extends to conducting insightful Data Analysis on real-world datasets and presenting findings through compelling visualizations. I take delight in the process of training Machine Learning and Deep Learning models, instilling programs with the intelligence to make informed decisions. Beyond development, I relish showcasing my work to the world by deploying models on web or app platforms. It's not just about coding; it's about crafting intelligent solutions that resonate with real-world challenges.

Education

Computer Engineer!

I completed my Computer Engineering from University of Engineering and Technology, Peshawar. I completed my engineering degree in the year of 2023 with 3.41 CGPA (out of 4). The major subjects that I studied during my 4 years are:

  • Computer Fundamentals
  • Computer Programming (C++)
  • Calculus
  • Object Oriented Programming (C++)
  • Probability and Statistics
  • Web Engineering (HTML, CSS, Javascript)
  • Data Structures and Algorithms
  • Linear Algebra
  • Operating Systems (C, Linux)
  • Computer Organization and Architecture
  • Systems Programming (C)
  • Signals and Systems (MATLAB)
  • Digital Signals Processing (MATLAB)
  • Database Management System (MySQL, SQL)
  • Data Analytics (Python, Statistics, Machine Learning)
  • Computer Security

Skills

Python

Conditional Programming in Python

Data Structures in Python

OOP in Python

Machine Learning

Supervised Learning Algorithms

Unsupervised Learning Algorithms

Scikit Learn

Recommendation Systems

Web Technolgies

Django

Flask

Dash

Data Analysis and Visualization

Numpy

Pandas

Matplotlib

Seaborn

Plotly

Power BI

Deep Learning

Neural Networks

Natural Language Processing

Computer Vision

Tensorflow

Pytorch

Keras

Generative AI

LLM Models (Gpts + Open Source)

LangChain

LlamaIndex

Prompt Engineering

RAG

Finetuning

Others

Data Structures and Algorithms

C++

C

MATLAB

SQL

Problem Solving




Projects

Advance RAG using llama2, LangChain🦜️, Chromdb

Made an advanced RAG system using open-source LLM and LangChain using the following techniques:
- Get a hypothetical answer from LLM for a query, combine both of them to get accurate and closely related answers/documents from the vector database.
- Generate queries related to the original query using LLM and then use these queries together to get more appropriate related documents from our vector database.
- Used Chromdb to get related documents from the vector database.
- Using this technique, we can score the retrieval of the documents according to the query that we have set. Basically, every document will have a score showing how much related it is to our query.

Customized Meal Plan Automation for Restaurant

- The plan was to automate the process of generating customized meal plans based on specific dietary requirements, such as calorie counts or dietary restrictions.
- Users can now interact with the system by requesting meal plans that fit their needs. For instance, they might ask for a meal plan that provides 400 calories, is vegetarian, or excludes gluten.
- To address this, we utilized a Retrieval-Augmented Generation (RAG) system. By ingesting menu cards into a vector database, and then when a user sends a query, it gets the relevant items from the menu and combines the retrieval with the query using a well-engineered prompt. This setup allows us to generate customized meal plans with items from the menu.
- The integration of LLM and RAG technology automates the creation of meal plans, providing users with personalized recommendations swiftly and accurately, enhancing the overall dining experience.

Sakura (FYP)

- Trained a model with an accuracy of 96% for four emotions detection using BERT and Deep Neural Networks.
- Collected data about different exercises that can enhance a particular emotion.
- Made a recommendation system for recommending exercises for mood enhancement.
- Deployed all these models/features on a web app via Streamlit.

ScrapeInflateForecast

- Used BeautifulSoup to scrape the latest inflation rate data from the official website.
- Performed detailed time series analysis to understand stationarity, seasonality, trend, autocorrelation, partial autocorrelation, and distribution.
- Optimized ARIMA to a mean square error of 2.88 and also trained SARIMA and Facebook Prophet on the data.

Human Face Detection and Recognition

- MTCNN (Multi-Task Cascaded Convolutional Networks): Used MTCNN for face detection. MTCNN is adept at detecting faces in an image and providing bounding boxes around them.
- ResNet (Residual Network): Used a pre-trained ResNet model for feature extraction to obtain facial encodings or embeddings. These encodings are numerical representations of facial features used for recognition.
- Euclidean Distance (L2 Norm): Compared the embeddings of newly detected faces with images of authorized individuals. A set threshold of 0.7 was used, indicating that if the similarity exceeded 70%, the individual would be granted authorization.

Experience

  • Artificial Intelligence Engineer
  • JMM TECHNOLOGY
  • 09/2023 - Present

A. Cryptocurrency Trading Bot Development:

  • - Trained models (XGBoost, Random Forest, Decision Tree) for forecasting Ethereum's closing value over the next 4 hours by using historical Ethereum data and market fear and greed data to enhance prediction accuracy. Selected the XGBoost model for the final implementation due to its lowest RMSE (Root Mean Squared Error) loss.
  • - Developed a Random Forest model to predict the best APR (Annual Percentage Rate) by analyzing other protocols' APRs, coin-specific APRs, and news sentiment extracted from cryptocurrency news data.
  • - Created a prediction pipeline that extracts real-time coin APR data, protocol APR data, news sentiment data, and market fear and greed data. The pipeline predicts APR based on the amount, duration, and token specified by the user.


B. Riyadh's Different Places Reviews Sentiment Analysis and Analytics:

  • - Analyzed 2.1 million reviews of different places in Riyadh to extract sentiments from the reviews and identify specific aspects being discussed. Determined whether the sentiment was positive or negative, helping to understand what people liked or disliked about various aspects. Achieved 50% cost savings by utilizing the OpenAI GPT-4 batch API for sentiment extraction.
  • - Conducted detailed analysis on the results to provide actionable insights. Identified top aspects that people liked about popular places and pinpointed negative aspects that needed improvement. This information was aimed at helping place owners understand and address specific issues.


C. Airport Guidance for Disable People (MVP):

  • - Trained and finetuned Yolov8 and Yolov5 models for the American Sign Language Detection System for Saudi Airport Assistance, ensuring the high accuracy of detection by collecting the real world data and hand labelling the data.
  • - Developed a system that can detect the American Sign Language and convert it into the text for the guidance of disable people in the airport.
  • - The chatbot was developed to answer those queries of the disable people.


D. Document QNA System

  • - Developed an End to End project using LangChain and OpenAI API to get Answers from the Document, uploaded by the User.


  • Machine Learning Engineer (Internship)
  • LAB-D TECH LTD
  • 07/2023 - 08/2023

Key Responsibilities:

  • -Developed and Optimized Human Face Recognition System for a School by Using MTCNN and RESNET, that can Recognize Authorize Person and Send Request to Admin whenever Unauthorized Person is recognized to make him Authorized by Entering Name.
  • -Optimized the Overall Recognition System by Imporving the Accuracy of Recognition and FPS Lag.