Swaraj Khan’s Portfolio

Publications

Automated Q and A Chatbot: Harnessing AI for Efficient Information Retrieval
2024
ICCMLAI
Pune, India

Developed a PDF-based Q & A chatbot application that efficiently retrieves answers using predefined content without AI or NLP.
Implemented PDF parsing techniques to extract and structure Q & A pairs for quick lookup and response.
Streamlined the system for high accuracy and low latency in response times, making it suitable for real-time applications.
Presented findings at the International Conference on Computer Science, Machine Learning, and Artificial Intelligence (ICCMLAI) in Pune, highlighting innovative methods for non-AI-based information retrieval systems.

Developed an AI powered chatbot for the internal teams to assist with log analysis during testing.
Using NLP, the chatbot helps employees understand log messages by providing suitable answers to their queries.

Built a model from patient database to predict heart stroke.
Data cleaning, data manipulation, handling missing values
Feature selection, confusion matrix
Data visualisation-Matplotlib: correlation matrix and bar graphs
Algorithms: Random Forest(0.945), Gradient Booster(0.941) & Logistic Regression(0.754)

GitHub Repo
Implemented a RAG chatbot using Gemini 1.5 Flash and Langflow, integrating advanced retrieval-augmented generation capabilities.
Utilized Chroma DB for efficient and scalable vector storage and retrieval of movie-related data.
Applied chunking and text splitting techniques for optimized data processing and context-aware response generation.
Engineered prompts and employed AI embeddings to enhance the chatbot’s accuracy upto 95% and relevance in answering queries about Constitution of India.

GitHub Repo
I created a GitHub repo for a project called Auto ML pipeline which automated machine learning tasks including image segmentation, LSTM prediction, and CSV data analysis.
This pipeline aimed to simplify the process of implementing machine learning models for various tasks, providing a seamless experience for researchers and practitioners alike.

GitHub Repo
Developed a Python-based NLP project analyzing “The Shawshank Redemption” script
Utilized various NLP techniques including sentiment analysis, word frequency analysis, and cosine similarity
Visualized character networks and sentiment trends
Conducted extensive text mining and data cleaning

GitHub Repo
I explored image classification using deep neural networks in the “Cat vs. Dog” project, assessing the performance of 2-Layer (72% accuracy) and L-Layer (80% accuracy) neural network models.
Detailed architecture, implementation, and performance metrics in a Jupyter notebook.
The future objectives involved experimenting with different architectures, delving into convolutional neural networks (CNNs), and expanding the model to tackle various image classification tasks.

GitHub Repo
Neural Style Transfer (NST) used a previously trained convolutional network, and built on top of that. The idea of using a network trained on a different task and applying it to a new task is called transfer learning.
Following the original NST paper, I used the VGG network. Specifically, I used VGG-19, a 19-layer version of the VGG network. This model had already been trained on the very large ImageNet database

Category	Skills
Data Science / Analysis	Neural Networks, Hyper Parameters Tuning, Convolution Neural Networks, Sequence Models, TensorFlow, Pandas, Numpy
MySQL / Database	Tables, Views, Queries, Indexes, Clauses, Conditions, Key, Aggregate Functions
Statistics / Concept	Conditional probability, Bayes Theorem, Probability Distribution Function
Power BI / Data Visualization	Sales Insights, Spotify Dashboard, Amazon Prime Video Dashboard