Explore my journey through AI, Machine Learning, and Backend Development. Each project represents a step forward in my technical evolution.
Built an intelligent job recommendation system that analyzes CVs and provides personalized job matches with career guidance. Implemented a sophisticated hybrid search approach combining BM25 for keyword-based extraction and semantic search for contextual understanding. Leveraged Qdrant vector database for efficient storage and retrieval of job embeddings. Enhanced search relevance using Reciprocal Rank Fusion (RRF) to merge results from multiple retrieval methods, followed by cross-encoder reranking for optimal job matching. Integrated Groq's Llama model via LangChain for generating contextual career advice and job insights. Implemented advanced job description processing with semantic chunking and metadata storage for scalable retrieval and recommendation accuracy.
Developed a deep learning model using CNNs to detect pneumonia from Chest X-rays as college major project. Enhanced model performance through data augmentation, achieving high accuracy for robust pneumonia detection. Used Grad-Cam for showing affected parts in x-rays. Developed a full-stack web application using Django as backend with React as the frontend.
Developed a mobile app where user can request and donate blood as college minor project. Implemented a feature to search for compatible blood donors using an integrated map and filters like city, Bloodgroup. Implemented other features like BMI calculation, JWT authentication.
Performed semantic vector embeddings for cleaned book dataset using Sentence Transformer to recommend top-k books based on user's query. Integrated T5 and Open AI(DeepSeek) models to generate contextual responses to users. Stored embeddings and metadata using PgVector and Supabase for efficient retrieval and RAG-based query resolution.
Built an AI-powered email assistant that reads, understands, and responds to emails using Gemini LLM via LangChain. Integrated Gmail API for secure retrieval of unread emails and automated inbox polling on a schedule. Implemented PDF parsing and information extraction from attachments, feeding relevant context to LLM for accurate replies. Leveraged LangChain's tool calling to invoke specific actions (e.g., data extraction, lookup, or follow-up triggers) based on email intent.
Built a PDF/text processing API that chunks documents semantically and stores them in Supabase with vector embeddings. Enabled smart search by implementing pgvector similarity search for retrieving relevant document sections. Developed RESTful endpoints with proper validation, error handling and async processing.
Performed Data Cleaning, EDA (Uni-Variate/Multi-Variate), Feature Engineering over raw car datasets. Used Random Forest Regressor, and XG Boost Regressor for comparing metrics and fine-tuned them for better results. Evaluated the performance of models using Dummy Regressor.
Developed a Fast API based application for predicting house price based on different features. Implemented 12-factors principle like Testcase using pytest, dockerized for easy deployment, followed Cookie-Cutter Data Science template and many more.
Exploratory Data Analysis (EDA) and visualization of the MPG dataset using Matplotlib, Seaborn, and Plotly. This project focuses on exploratory data analysis (EDA) and interactive visualizations of the classic MPG (Miles Per Gallon) dataset. The goal is to uncover meaningful patterns and relationships between various features of automobiles such as horsepower, weight, origin, cylinders, and fuel efficiency (mpg).
Classic machine learning project implementing various classification algorithms to predict iris flower species with comprehensive data visualization and analysis.
Deep learning project using MNIST dataset to classify handwritten digits with high accuracy using neural networks and CNNs, featuring interactive visualizations.