Build real-world AI & ML applications.

Build an end-to-end text classification pipeline — data cleaning, TF-IDF features, logistic regression vs. BERT, evaluation metrics, and Flask API deployment.

• NLP
• scikit-learn
• Transformers
• Flask

Train a CNN from scratch on CIFAR-10, then fine-tune a pre-trained ResNet. Compare accuracy, training time, and deploy with Gradio.

• Computer Vision
• PyTorch
• ResNet
• Gradio

Full EDA, feature engineering, and model comparison (Linear Regression, Random Forest, XGBoost). Includes cross-validation and hyperparameter tuning.

• Tabular ML
• pandas
• XGBoost
• Feature Eng.

Handle class imbalance with SMOTE, build ensemble models, create a Streamlit dashboard, and design an automated retraining pipeline.

• Tabular ML
• Imbalanced Data
• Streamlit
• MLflow

Fine-tune T5/BART on a summarisation dataset, evaluate with ROUGE metrics, and serve via FastAPI with streaming responses.

• NLP
• Transformers
• Fine-Tuning
• FastAPI

Train YOLOv8 on a custom dataset, build a real-time inference pipeline with webcam input, and deploy to an edge device.

• Computer Vision
• YOLOv8
• Real-Time
• Edge Deploy

Build an Airflow-orchestrated pipeline that ingests from APIs, transforms with dbt, loads into a data warehouse, and includes data quality checks.

• Data Engineering
• Airflow
• dbt
• PostgreSQL

Build a Retrieval-Augmented Generation chatbot with LangChain, vector database (Chroma/Pinecone), document chunking, and citation tracking.

• GenAI
• RAG
• LangChain
• Vector DB

Fine-tune an open-source LLM (Llama/Mistral) with LoRA/QLoRA on a domain-specific dataset. Evaluate with perplexity, MMLU, and human preference.

• GenAI
• LoRA
• PEFT
• Hugging Face

Build a monitoring system that detects data drift, model degradation, and prediction anomalies. Includes alerting and automated retraining triggers.

• MLOps
• Evidently AI
• Grafana
• Docker

Implement collaborative and content-based filtering, matrix factorisation, and a hybrid recommender. Serve recommendations via a REST API.

• RecSys
• Matrix Factorisation
• Python
• API

Build an autonomous agent with tool-use capabilities — web search, code execution, database queries. Implement ReAct loop, memory, and guardrails.

• GenAI
• Agents
• Tool Use
• Guardrails

Forecast stock prices or energy demand using ARIMA, Prophet, and LSTM. Compare statistical vs. deep learning approaches with proper backtesting.

• Time Series
• Prophet
• LSTM
• Backtesting

Train a reinforcement learning agent to play Atari games using DQN and PPO. Visualise training curves, reward shaping, and policy behaviour.

• RL
• DQN
• PPO
• Gymnasium

Build a complete ML platform — data ingestion, feature store, model training, A/B testing, monitoring, and CI/CD. The ultimate portfolio piece.

• MLOps
• Full Stack
• Kubernetes
• CI/CD
• Feature Store