Projects

This is a collection of some personal and professional projects I’ve worked on. More projects available on my GitHub.

---

GenAI & LLM Projects

Image Similarity Search using CLIP Model

Code: ImageSimilaritySearch
• Leveraged OpenAI’s CLIP model to build an image search application that allows users to search for images based on text descriptions.
• Generated embeddings for images using CLIP, computed cosine similarity scores between text and image embeddings, and ranked results. Built with Streamlit for an interactive UI.

Generate Survey Questions on given Persona and Goal

Code: MetaformsAISurvey
• Streamlit application that generates goals and survey questions using OpenAI’s GPT model.
• Users can input a persona and domain to generate goals, then create survey questions across multiple iterations with organized, expandable output sections.

Context Aware Chatbot with Unstructured/Structured Data

Code: Custom AI ChatBot
• Utilized SQLAlchemy as the data connector for SQL and BigQuery databases, generating SQL queries from natural language input using llama-index.
• Created analyzer charts and natural language responses based on SQL output, with OpenAI APIs and VectorDB for chatbot interactions with unstructured data.

LLM Customised ChatBot (RAG Application)

Code: LLM-CustomisedChatBot
• Simple RAG application built with Streamlit for conversational AI on custom knowledge bases.

Collaborative Filtering & Association Rule Mining

Code: RecsysAssociationPy
• Recommendation system using Collaborative Filtering, Apriori Algorithm, and FP-Growth for Association Rule Mining.

---

Computer Vision & Deep Learning Projects

Queue Detection & Length Estimation using Deepstream

Code: QueueDetection DS App
• Developed a Deepstream application to detect queues and estimate queue length. PeopleNet model for detecting people and convex hull for finding the Queue vector (y = mx + c).

XML to KITTI Format Converter

Code: xml2Kitti
• Python script to convert XML annotation format to KITTI format for object detection model training.

Multi-Label Classification on Street Fashion Dataset (Modanet)

• Performed object detection, segmentation and polygon detection to classify clothes in images across 13 meta fashion categories using Keras and OpenCV.

---

ML Engineering & DevOps Projects

Flask ML API: Salary Predictor App

Code: SalaryPredictApp
• End-to-end ML deployment: trained a Linear Regression model with scikit-learn, built a Flask web app for serving predictions via REST API.

Setting Up JupyterHub using Docker

Repo: TLJH_Docker
• Set up JupyterHub on a local server using Docker and docker-compose with multi-user basic authentication. Detailed guide: Blog

Record Matching for Healthcare & Retail

Code: RecordMatching
• Patient and entity record matching system for Hospital and Retail domains.

Kailash Hospital Website

Code: kailash-hospital-website
• Official website for Kailash Multi Speciality Hospital & E.N.T. Center, Banda, Uttar Pradesh.

---

Publication

Automatic Generation of Chest X-Ray Medical Imaging Reports using LSTM-CNN

International Conference on DSMLAI, Windhoek, Namibia (Aug 2021)
• Developed methods to automatically generate text-based reports for medical images, helping medical professionals generate reports more precisely and in less time.

Affordable AI at the Edge on NVIDIA’s Jetson Ecosystem

Major Project Thesis, IIIT Naya Raipur (July 2020)
• NVIDIA’s End-to-End Tech stack for Intelligent Video Analytics on Jetson Devices. Transfer learning toolkit for training and deploying models with Deepstream.

---