Mohsin Raza

End to End Multimodal Claim vs Opinion Classification for TikTok Videos

A production level modular data science project aims to predict generous tippers for taxi drivers.

Automated candidate ranking for HR: applies NLP, skill matching, cosine similarity, and LLMs to identify the top 10 profiles matching a job description.

Instruction fine-tuning of the DeepSeek Large Language Model using Unsloth’s FastLanguageModel and the alpaca-gpt4 dataset, enabling fast, memory-efficient adaptation of a foundation LLM for high-quality instruction following and real-world AI applications.

Predicts residential house prices using structured data and a PyTorch-based deep learning model, built with production-ready ML engineering practices including testing, Dockerization, CI, and API-based inference.

End‑to‑end ML project predicting loan approvals using 9 applicant features. Data is processed, modeled with Random Forest and GridSearchCV, deployed via FastAPI and Uvicorn, with a simple frontend, containerized in Docker, tested with pytest, and automated through CI/CD.

This project analyzes and predicts taxi fares estimate fares in advance using Regression Analysis. Conducted EDA, hypothesis testing, to identify key variables. Developed ML models (Random Forest, XGBoost) with GridSearchCV for hyperparameter tuning to predict generous tip giver accurately.

Built a payment delay prediction system using the "Default of Credit Card Clients" dataset. Leveraged a Random Forest classifier to predict next-month default. Deployed the model via FastAPI with Uvicorn server for fast, scalable API access. MongoDB Cloud stores client data and prediction results securely.

A simple context retrieval RAG (Retrieval-Augmented Generation) pipeline involves several steps: data indexing, retrieval, and generation. First, the data is loaded, split into smaller chunks, then embeddings are created for the chunks and stored in a vector database. When a query is received, retrieves the most relevant chunks from the database.

This project focuses on fine-tuning Meta’s LLaMA 2 model to develop a domain-specific medical chatbot capable of understanding and responding to patient and clinician queries with high accuracy. Leveraging parameter-efficient fine-tuning techniques—LoRA and QLoRA the project ensures resource-efficient training while maintaining high performance.

This project aims to build an AI-powered Legal Advisor that leverages natural language processing and vector search technology to provide users with legal guidance based on authoritative legal texts.

A secure backend user authentication system using FastAPI, JWT, and MongoDB Atlas. It supports user signup, login, JWT-based session management, and CRUD operations for user profiles. Passwords are hashed for security. MongoDB Cloud ensures scalable data storage with fast and reliable access.

For this purpose we used Seoul dataset. Seoul is a company offers various bike rental options for tourists and residents. Seoul Public Bike, is an unmanned rental system that can be used anywhere, anytime by anyone, designed to address traffic congestion, air pollution, and high oil prices in Seoul, enhancing the quality of life for citizens.

This FastAPI app includes a GET endpoint at / and a POST endpoint at /send to receive user input. It retrieves data from a simulated database implemented as a Python dictionary. The app uses Pydantic's BaseModel to validate input data and ensure type safety.

This is a sample repo for git and github practise

Building a CNN model using SVHN dataset

CNN model on facial emotion recognition using FER 2013