Mohsin Akbar Email & Phone Number

United States

- Create and curate high quality datasets for Supervised Fine Tuning (SFT).- Evaluate RLHF datasets for LLM responses, and fine tuning with human feedback.- Fine tune Large Language Models- Stress testing LLMs for errors and break points.- Training LLMs to write high quality code with logical reasoning to carry out tasks.- Training LLMs to generate complex visualisations using all sorts of user queries.- Engage in meetings and discussions to improve the quality of LLM responses- Evaluating different loss buckets of different LLMs

Islamabad, Islāmābād, Pakistan

- Write clean, efficient and complex SQL queries to carry out data analysis- Write stored procedures to generate travel days reports to charge clients.- Leverage the power of Python to streamline and automate data cleaning and data preprocessing.- Help test different data parsers and data verification as part of Quality Assurance.- Help verify and detect data anomalies and fix them.- Leverage Python and SQL to find out bugs in the system.- Performed streamlined data manipulation using Python, saving an average of 3-4 hours per day and improving data accuracy by 30%.- Automated the report generation process using Python, saving an average of 1 hour per report cycle and reducing the report generation time by 90% by developing a frontend that generates reports away from a user’s click.- Using statistical analysis and visualizations to create reports.- Work closely with cross-functional teams to ensure clean and valid data in the system.- Take part in meetings to help discuss data and system issues and collaborate with clients and business analysts to improve the system.- Stay updated with the latest tools and technologies in data and AI world.

Pakistan

- Developed XGBoost classifier that detected 98% of financial frauds, significantly surpassing the previous 80% accuracy and safeguarding the bank millions in losses by doing feature engineering and feature selection leveraging ML techniques.- Analyzed the data of “Fortune 500” companies and performed exploratory data and statistical analysis to identify patterns and trends to recognize areas for profitable growth and leveraged machine learning to forecast future profits.- Analyzed dataset of machine sensor readings and failure histories to identify patterns and trends using statistical methods and machine learning algorithms. Conducted failure analysis to quantify the potential savings for the client, and provided recommendations on how to optimize their maintenance process and reduce costs in millions.- Analyzed the data of factory workers, found different parameters related to the overall efficacy of works, and increased the previous R2-score of predicting their efficacy from 29% to 67%.- Developed Random Forest classifier with 88% accuracy, trained & evaluated on a large dataset of tweets to detect hate/offensive tweets by leveraging different NLP techniques, built Streamlit web app to facilitate reporting of flagged tweets.- Completed a total of 50+ small and large projects for international clients and promoted to level 2 within a single year by being persistent, consistent and eager to learn and grow.

United States

- Devised user-user recommendation system based on collaborated filtering techniques, explaining the working mechanism of recommendation systems. The recommendation system helped increasing the overall user engagement by 12%.- Conducted hypothesis testing on crop trial data to identify treatment effects of three different drugs, leading to increased yield of14%. This also helped in overall understanding of applying statistical tests to real world scenarios.- Developed several video tutorials in Jupyter Notebook, enabling students to learn key concepts of data science more efficiently & comprehensively, which helped in increasing the overall revenue in course sales.