S2Cool Training Module

A Project of the S2Cool Initiative

4-Day Intensive Training

Applied AI & Data Science for
Environmental Sustainability & Cooling Solutions

Build an LLM-powered RAG assistant and deploy Computer Vision models. A hands-on workshop dedicated to solving climate challenges.

Format 4 Days × 6 Hours/Day (Lecture + Hands-on Lab)

Daily Structure 2 Lecture Options + 3h Lab Block

Audience Analysts, Engineers, Researchers

Prerequisites Python, ML Basics Command Line + Git

Learning Objectives

By the end of Day 4, participants will be able to:

Build an LLM-powered RAG assistant for climate/sustainability/cooling knowledge and reporting.

Train and evaluate computer vision models for environmental monitoring (heat islands, deforestation, land cover).

Productionize models with MLOps/LLOps: Experiment tracking, model registry, CI/CD, monitoring, containerization, scalable inference.

Write clean, testable AI software: Modular pipelines, APIs, packaging, performance basics.

Complete Timetable

Detailed breakdown of lectures and hands-on labs.

Day 1: LLMs & Generative AI

Climate Data Science Focus

Instructor: Dr. Muhammad Moazam Fraz

09:00 – 10:30	Lecture 1 (90m) LLMs, climate domain framing, RAG foundations
10:45 – 12:15	Lecture 2 (90m) Embeddings, retrieval, evaluation, safety/grounding
13:30 – 16:30	Lab (180m) Build Climate+Cooling RAG assistant v1 Hands-on implementation of the concepts learned.

Day 2: Computer Vision

Environmental Monitoring

Instructor: Dr. Nazia Pervaiz

09:00 – 10:30	Lecture 1 (90m) Remote sensing + CV tasks for sustainability
10:45 – 12:15	Lecture 2 (90m) Segmentation/classification, metrics, robustness
13:30 – 16:30	Lab (180m) Heat island / land cover / deforestation model + inference

Day 3: MLOps/LLOps & Deployment

From Prototype to Production

Instructor: Dr. Fahad Satti

09:00 – 10:30	Lecture 1 (90m) MLOps lifecycle, experiment tracking, reproducibility
10:45 – 12:15	Lecture 2 (90m) CI/CD, model registry, monitoring, scaling (Ray/Spark)
13:30 – 16:30	Lab (180m) Containerize + deploy APIs + MLflow registry + basic monitoring

Day 4: Programming for AI

Clean Code, Pipelines, APIs

Instructor: Dr. Sahar Arshad

09:00 – 10:30	Lecture 1 (90m) Clean architecture for AI systems + testing strategy
10:45 – 12:15	Lecture 2 (90m) Data pipelines, APIs, packaging, performance basics
13:30 – 16:30	Lab (180m) Refactor to production-ready repo + tests + final demo

Meet the Trainers

Dr. Muhammad Moazam Fraz

Principal Investigator

Dr. Nazia Pervaiz

Instructor

Dr. Fahad Satti

Instructor

Dr. Sahar Arshad

Instructor

Integrated Capstone Project

"CoolSmart Climate & Cooling Intelligence Platform"

A deployable prototype that helps decision-makers identify heat-risk zones, understand sustainability drivers, and generate evidence-grounded cooling interventions. The project evolves across all 4 days.

LLM / GenAI (Day 1)

Climate & cooling RAG assistant: Summarize climate risks and cooling strategies.
Generate sustainability briefs with citations.
Q&A over IPCC/standards/cooling guidelines and local notes.
Outputs structured JSON for downstream dashboards/workflows.

Computer Vision (Day 2)

Environmental monitoring module: Detect urban heat island risk proxies (LST + built-up/vegetation).
Optional extension: deforestation or land cover change.
Outputs masks/tiles + risk scores for neighborhoods/areas.

MLOps & Production (Day 3 & 4)

Experiment tracking + model registry (MLflow).
FastAPI microservice endpoints:
/cv/predict (tile → mask + stats)
/rag/ask (question → cited answer)
Containerized deployment + telemetry logs.
Optional scalable batch inference (Ray).
Clean architecture, tests, packaging, configs.
Performance tuning (caching, batching) + Final demo case study.

Open-Source Tech Stack

Python & Jupyter pandas & numpy PyTorch Hugging Face LangChain/LlamaIndex OpenCV MLflow Docker FastAPI

Applied AI & Data Science for
Environmental Sustainability & Cooling Solutions

Learning Objectives