EvalBench - Python library for Smart LLM Evalution

EvalBench is a package for evaluating LLM outputs across 15+ metrics like response quality, retrieval accuracy, hallucination, etc. Now with agentic workflows: simply describe your goal, and EvalBench automatically plans and runs tailored evaluation, analysis, and suggests improvements.

MLOps Toolkit

A modular pipeline integrating MLflow, Hydra, Flask, Docker, and GitHub Actions, deployed on AWS EC2 with monitoring via Prometheus, Grafana, and CloudWatch, including a retraining workflow.

MeetTrack - Your Helpful Meeting Assistant

A smart meeting assistant that handles post-meeting administrative task like producing meeting notes that capture visual content, accolades and recognition along with key meeting points. A standout feature is a chat feature for real time interaction with the meeting content.

Smart Ad Generation with AI

In the ever-evolving landscape of digital marketing, the ability to create engaging and targeted advertisements is paramount. This project aims to transform this process by developing an AI driven system that generates both textual ad copy and visual ad images seamlessly.

Radiology AI Assistant and Chatbot

An application that takes a chest X-Ray image and generates a medical report using transformer model. Along with that, a chatbot facility is available that can look at the generated report and answer user questions based on the report.

LLM Finetuning, LLM Quantisation and Retrieval Augmented Generation

- Llama FInetuning - LLM Qauntisation using LoRA and QLoRA - RAG using Mistral

Prompt Engineering

Collection of prompt engineering techniques, prompt templates, and prompts generated using OpenAI.

SafeDose - Monitoring substance abuse using ML

‘SafeDose’ is a tool driven by data science algorithms to assist in identifying the different kinds of substance abuse occurring in the country as well as the overcoming the challenges of lacking medical documentation that leads to failure in identifying the cause of the drug misuse episode.

Lung Cancer Detection using Deep Learning and Image Processing

A Deep Learning based solution to detect and classify lung cancer nodules as malignant or benign. The custom 3-dimensional Deep Convolutional Neural Network (3D DCNN) architecture detects the nodule candidate from the CT scans and further classifies it as benign or malignant.

SafeSpace - Giving Walkers Safer Neighborhoods

An android app that uses the number of people in a particular area to suggest safe spaces to users when they find themselves in unfamiliar locations. The app stars three features: crowd density estimation, location and direction to open establishments, and informing contacts via a distress signal.

Forecasting Sensor Measurements in Smart Air Quality Monitoring System

Analyze factors affecting air quality using sensor data collected from air quality monitoring systems. Use this sensor data to anticipate future trends in temperature and CO concentrations so as to allow appropriate preventive measures.