Clean raw YouTube comments, vectorise them with BOW/TF-IDF/Word2Vec, and train KNN and Logistic-Regression models to spot toxicity. Along the way you’ll learn smart data splits, hyper-parameter tuning, and how to read accuracy, F1, ROC-AUC, and confusion-matrix results with confidence.

• Clean raw text and create vectors with BOW, TF-IDF, Word2Vec
• Train & tune KNN and Logistic Regression on toxic-comment data ⎯ distance metrics, regularization, hyper-parameter search

• Judge models using Accuracy, F1 and understand bias-variance & interpretability

Build perceptrons and multilayer networks from scratch, then implement them in PyTorch. You’ll master forward/back-prop, optimisers (SGD → Adam), regularisation tricks like dropout and batch-norm, and keep track of experiments in TensorBoard—skills that transfer directly to more advanced deep-learning work.

• Build perceptrons → multilayer networks; grasp forward/back-prop & activation choices
• Leverage optimizers (SGD, Adam, AdaGrad…), weight init, dropout & batch-norm for stable training

• Use PyTorch: data loaders, custom trainer loop, TensorBoard tracking, and inference pipelines
• Hands-on: build classifiers in PyTorch

Move beyond static text into sequences: train RNNs, LSTMs, GRUs, and attention-augmented variants before unpacking the Transformer. Hands-on labs with BERT, GPT-2, and Llama 2/3 teach you tokenisation, positional embeddings, fine-tuning, and latency metrics, setting the stage for modern language-model engineering.

• Move from vanilla RNN/LSTM/GRU to attention-powered networks
• Decode the Transformer architecture—tokenizers, positional embeddings, multi-head attention
• Learn about Decoder and Encoder Only Transfromer

• Fine-tune BERT, GPT-2, Llama 2/3; measure latency with TPS & TTFT; master prompting basics
• Build Named Entity Recognition (NER) and Text Generation Models

Combine LLMs with vector databases to build RAG pipelines that retrieve, rank, and inject context for more grounded answers. You’ll also experiment with agentic patterns that let multiple models collaborate on multi-step tasks, plus tips for boosting retrieval quality and throughput.

• Build a full RAG pipeline: embed docs, store in vector DB, retrieve & rank for context-aware answers
• Optimize retrieval quality and latency; integrate AI agents to orchestrate multi-step tasks

• Build a workflow with AI Agents

Turn a trained model into a real product: wrap it in a Flask-Gunicorn API, containerise with Docker, push to Google Artifact Registry, and deploy on Cloud Run with autoscaling. Finish by fronting the service with API Gateway, securing it with keys, monitoring usage, and tuning cost-vs-latency trade-offs.

• Wrap models in a Flask + Gunicorn API and profile CPU / GPU usage
• Containerize with Docker, push to Google Artifact Registry, and "Deploy" on Cloud Run with autoscaling

• Expose your service with API Gateway, and secure with keys
• How to optimize your AI Service / API

Think like an AI engineer solving customer problems: source data, frame hypotheses, and iterate quickly. Projects include voice-of-customer analysis, résumé parsing, moderation bots, and chat assistants—each emphasising experiment design, cost control, and deliverables that matter to stakeholders.

• Think like an AI engineer: scoping, data sourcing, and rapid experiment loops
• Deliver voice-of-customer analytics, resume parsing, moderation bots, and chat assistants end-to-end

• Work on cost optimisation and model-compression strategies

Dive into CNNs, receptive-field maths, and object-detection workflows with YOLO. You’ll practise image augmentation, evaluate with mAP, and contrast CNNs with Vision Transformers and hybrid CNN+ViT setups—preparing you for any modern vision stack

• Master CNN layers, pooling, and receptive-field maths; design custom image-classifiers
• Train YOLO for object detection, evaluate with mAP, and apply targeted image augmentation

• Explore Vision Transformers (ViT), hybrid CNN+ViT, and vision-language models (VLM)