JazilKalim
JazilKalim
JazilKalim
I designed and implemented LLM solutions to automate and enhance specific aspects of a car dealership’s operations to improve customer experience and operational efficiency by implementing pre-trained large language models (LLMs) for various tasks.
Project Link:
https://github.com/jazilkalim/Analyzing-Car-Reviews-with-LLMs-
I developed a dashboard for the AI Team at Afiniti using Apache Superset. This dashboard revolutionized the way we review hundreds of alerts and scripts. It offers a new approach to assess the AI model’s performance in production, allows us to monitor production statistics using statistical tests, and automatically sets intelligent alarm thresholds to raise critical problems beforehand.
This project involved building a real estate price prediction model using regression that allows users to interact with the model and get real-time predictions, such as estimating a house’s sale price.
In this project, I utilized Tableau to analyze trends in Foreign Direct Investment (FDI) data. I explored the data to identify sectors receiving the highest and lowest levels of funding, along with those experiencing the most significant growth and decline. This analysis allowed me to uncover investment trends, group sectors based on commonalities, and identify variations within each sector. Furthermore, I employed clustering techniques to identify groups with similar FDI patterns and forecasted future trends. To effectively communicate these findings, I designed an interactive Tableau dashboard that allows users to explore the data and gain insights into FDI patterns.
Project Link:
https://github.com/jazilkalim/Foreign-Direct-Investment-Insights
Fueled by the global surge in interest for electric vehicles (EVs), my final year project focused on designing and simulating a battery pack for Pakistan’s first domestically produced EV, an initiative undertaken by the DICE foundation. To determine the vehicle’s battery requirements, I collected data on Karachi’s road conditions to create a realistic vehicle speed profile. Then I utilized Simulink to model and simulate the vehicle’s powertrain and battery pack. By analyzing the resulting output graphs, we were able to extract key findings and identify any abnormal responses within the system.
Furthermore, recognizing the crippling impact of Pakistan’s energy deficit and import bill, the project delved into data on the nation’s energy situation. We used this data to advocate for the development of an electric vehicle policy, promoting the use of EVs as a potential solution.
It was also mentioned in multiple news forums when the car was launched
This project involved a deep dive into stock market prediction using deep learning. I began by retrieving stock data from Yahoo Finance and Tiingo. To ensure consistency across different stocks, I resampled the data into various timeframes like quarters, months, and weeks. Next, I constructed moving windows to analyze how stock prices fluctuate over time. This analysis included measuring volatility, calculating rolling means (average price over a specific window), and visualizing the performance of different stocks using subplots (multiple plots within a single figure).
Following this data pre-processing stage, we built a deep learning model to identify patterns in the historical data. This model was then used to predict future stock prices.
Project Link:
This project used computer vision to identify five sports players. I collected images using Fatkun, processed them with OpenCV, and detected faces with Haar cascade classifiers. These faces were cropped to create a dataset for training machine learning models. I experimented with SVMs and CNNs, fine-tuning hyperparameters to select the best model for player classification.
Project Link:
https://github.com/jazilkalim/Sports-Celebrity-Face-Classifier
Sowing Success, using ML to help farmers select the best crops:
Traditional soil health assessment is costly and time-consuming. This project aims to leverage this data by building multi-class classification models. Models will predict the most suitable “crop” for a given soil profile, ultimately identifying the single most important feature for accurate predictions. This approach has the potential to streamline crop selection for farmers and optimize their agricultural practices.
Project Link:
https://github.com/jazilkalim/Predictive-Modelling-for-Agriculture
Exploring Nobel Prize Laureates (1901-2023):
This project delves into the prestigious Nobel Prize.Through data analysis, I uncovered insights about the prize-winners. I analyzed trends in Gender and Nationality about which gender and birth country are most represented among Nobel laureates, which decade saw the highest proportion of US-born winners compared to the total awardees and across the Nobel categories, which decade saw the highest percentage of female winners and Who was the first woman to receive a Nobel Prize and in which field.
Project Link:
https://github.com/jazilkalim/Nobel-Prize-Winners-Analysis
This project explores building a book recommendation system. It utilizes two key techniques: collaborative filtering and Pearson correlation. Pearson correlation is then employed to assess the strength of the relationship between the target book and those of their nearest correlated book vectors. Collaborative filtering leverages the power of user ratings. The system identifies users with similar book rating patterns (using K Nearest Neighbors with cosine similarity) to the target user. These nearest neighbors act as a reference point for recommendations.
Books with ratings most closely aligned with the target user’s preferences (i.e., having the lowest cosine distance) are then presented as the top 5 recommendations. This approach aims to suggest books that are likely to resonate with the user based on their past reading choices.
Project Link:
https://github.com/jazilkalim/BookRecommendationSystem
This projects builds a Streamlit web app to predict penguin species. Users can either upload a CSV file containing their penguin data or manually enter features like island, sex, and body measurements. The app preprocesses the data, including encoding categorical features like island for the model. A pre-trained classification model is then loaded and used to predict the penguin species based on the user input. The app displays both the predicted species (Adelie, Chinstrap, or Gentoo) and the prediction probabilities, providing users with insights into the model’s confidence level for each prediction.
Project Link:
https://github.com/jazilkalim/Penguins-Type-Prediction-Using-Streamlit-and-Heroku/tree/main