AI & MLOps Course

• Essential Tasks in Computer Vision
• Convolutional Operation - kernels, padding, feature maps
• Pooling Operation
• CNN for Image Classification
• Transfer Learning – Backbone of modern CV
• Residual Connection, Batch Normalization for training deeper networks
• Depthwise Separable Convolution and Xception
• Object Localization and Detection Algorithms - YOLO
• Image Segmentation - UNet and DeepLab
• Computer Vision in Operation – Best practices and tools

• Cloud Computing foundations
• Need for cloud computing
• Introduction to AWS and Azure systems
• Cloud9 and EC instances
• AWS sage maker
• Cloud Computing Tools - AWS ML Tools; Google Cloud; IBM Watson

• MLOps & Version Control
• What is DevOps and MLOps
• What is CI and CD why do we need them ?
• What is version control - git basics - clone, pull, push, brach, commit, merge conflicts
• Docker - Need for containerisation of applications
• How to setup a docker container
• How to mount volumes and enable ports within docker
• How to write a docker compose file
• CI & DVC - Github actions - Data Version Control - DVC Pipelines
• MLOps Tools and ML Deployment - Jenkins - Dags Hub - Weights and Biases or Mlflow
• Basics of python flask for web deployment

• Essential Tasks in NLP
• Data Preprocessing for Language Models
  1. Text Vectorization Layer
  2. Standardization, Vocabulary Indexing
  3. Embedding Word Vectors
  4. TF-IDF
• Tokenization fundamentals, Byte Pair Encoding
• Bag of Words Model and Sequential Models
• Attention Mechanism
• Transformer Encoder for Language Comprehension/Understanding tasks
• BERT Models

• Representation Learning: The core of modern AI
• Decoder only GPT class of models for Language Generation tasks
• LLMOps – Tools, Platforms (incl, but not limited to LangChain, OpenAI API)
• Prompt Engineering, RAG, LoRA (and variants)
• Generative Adversarial Networks, Diffusion Models for image generation
• Autoencoders for representation learning, pre-training
• Reinforcement Learning through Human Feedback
• Guardrails and processing of LLM outputs

• Parallel Programming Paradigms: OpenMP
• Scoping Variables and race conditions
• Parallel Algorithms using openMP: kmeans; histogram computation; Basic problems on loop interchange and blocking
• Parallel Programming Paradigms: MPI
• Cuda
• Submitting jobs on Cluster
• Python Parallel Programming - Multi Threaded vs Multi Processor in python - GIL -
• Introduction to PyOMP - Introduction to PythonMPI

• GPU's in ML - Introduction to GPU architecture - Tensor cores and Mixed precision -
• Distributed computing with tensorflow (GPU)
• Parallel ML Libraries: Dask, Horovod, Ray, RAPIDS
• Introduction to cuDF, cuML, cuPY

Project Overview:
This project aims to develop a robust solution for detecting financial fraud using MLOps methodologies. It involves analyzing historical transaction data, customer behavior, and relevant factors to identify and prevent fraudulent activities in the financial domain. The financial sector faces a growing threat from financial fraud, driven by increasingly sophisticated tactics used by fraudsters. Traditional fraud detection methods are struggling to keep up with these evolving schemes. This project addresses the need for advanced fraud detection solutions capable of adapting to emerging threats in real-time.

Project Overview:
A Customer Conversational Intelligence Platform is a system that employs advanced technologies, including machine learning and natural language processing, to analyze and make sense of customer interactions across various communication channels such as chatbots, call centers, emails, and social media, that modern businesses accumulate. This project seeks to utilize this data goldmine to provide businesses with a competitive edge in customer service. Here, we will develop a platform that harnesses the power of machine learning to analyze vast amounts of customer interaction data. The aim is to derive actionable insights from these interactions, optimize customer service processes, and enhance overall customer experience.

Project Overview:
This project aims to predict the demand for delivery drivers in specific regions and times, leveraging MLOps methodologies. By analyzing order requests, driver activity, and related parameters, the goal is to optimize delivery charges, ensuring consistency and minimizing customer drop-offs. The unpredictable nature of delivery charges, primarily due to driver unavailability, often results in increased prices and subsequent customer dissatisfaction. This project seeks to bridge this gap by forecasting driver demand, thereby streamlining delivery pricing.

Project Overview:
The field of finance can be complex and overwhelming for individuals seeking personalized financial advice. In order to make informed decisions regarding investments, retirement planning, budgeting, and financial products, individuals often require guidance from financial experts. The aim of this project is to develop an Intelligent Financial Advisor powered by a Large Language Model (LLM) to provide personalized financial advice and guidance to individuals. By leveraging NLP and machine learning techniques, the Intelligent Financial Advisor will assist users in making informed financial decisions and achieving their financial goals.

Project Overview:
The goal of this project is to create an automated Search Engine Optimization (SEO) tool using ChatGPT, an AI-based chatbot system. The tool will use natural language processing (NLP) and machine learning (ML) algorithms to analyze website content, identify SEO issues, and provide recommendations for improvement. The tool will help website owners and SEO professionals to optimize their website's content and improve search engine rankings more efficiently and effectively.

Project Overview:
In the domain of scientific question answering, validating answers and providing accurate feedback is critical for effective learning. The goal of this capstone project is to develop an automated answer validation system using a Siamese text similarity model. The system will compare student responses with the correct answer and distractors to determine the level of correctness and provide appropriate feedback. The automated answer validation system for science question answering will benefit educators and students in science-related subjects. It will streamline the assessment process, reduce manual effort, and ensure consistent evaluations, leading to improved learning outcomes in the science domain.

Project Overview:
A GAN is a popular model for unsupervised machine learning where two neural networks — a generator and a discriminator, interact with each other. The generator generates images out of random noise it takes as input; The discriminator detects whether these generated images are fake or real (by comparing them to the images in a dataset). This process continues for several epochs until the discriminator loss between fake and real achieves its minimum. As the loss reaches the minimum, the generator becomes sufficiently skilled in generating images similar to those in the original dataset. AI-driven innovation with GANs has many applications in creative industries such as design. Be it architectural design, landscape design or interiors design, the possibilities are endless. Such generated designs have the potential to drive rapid growth and profits in the design industry. The goal of this project is to generate realistic new interior room designs by training a GAN network on the IKEA Interior Design Dataset.

Project Overview:
Crop losses due to diseases are a major threat to food security every year, across countries. Conventionally, plant diseases were detected through a visual examination of the affected plants by plant pathology experts. This was often possible only after major damage had already occurred, so treatments were of limited or no use. Recently, access to smartphone based image capturing has highly increased amongst farmers and agriculturists. This has led to the successful adoption of plant disease diagnostic applications based on deep learning techniques. This is of immense value in the field of agriculture and an excellent tool for faster identification and treatment of crop diseases. It holds key importance in preventing crop based food and economic losses. The goal of this project is to build a convolutional neural network or to use transfer learning and develop a plant disease identification tool.

Project Overview:
Worldwide, obesity has nearly tripled since 1975. In 2016, more than 1.9 billion adults, 18 years and older, were overweight (WHO sources). In such a situation, documenting dietary caloric intake is crucial to manage weight loss. Food image segmentation is a critical and indispensable task for developing health-related applications such as automated estimation of food calories and nutrients as a means for dietary monitoring. One of the challenges in this area is the improvement of accuracy in dietary assessment by food image analysis. However, how to derive the food information (e.g., food type and portion size) from food images effectively is a challenging task and an open research problem. In this project, participants are expected to make a model that can segment the food components present in an input food image and build an application that can predict the food class and the food portions from it.