606363 - Computing Systems for Data Science and AI

Course Overview

This comprehensive 16-week course explores the cutting-edge computing systems powering modern data science and artificial intelligence. From transformer architectures and large language models to AI agents, multi-agent systems, and diffusion models, students will gain both theoretical understanding and hands-on experience building intelligent systems.

What You'll Learn

Transformer architectures, LLMs, Mixture of Experts, AI Agents, ReAct, Multi-Agent Systems, Model Context Protocol, Large Action Models, Diffusion Models

What You'll Build

NER systems, RAG pipelines, AI Agents with LangGraph, Multi-Agent workflows, LAM applications, and generative AI projects

Technologies

Python, PyTorch, Hugging Face, LangChain, LangGraph, OpenAI API, Ollama, Playwright

Who Should Take This Course?

This course is designed exclusively for AI Students at University of Petra

Students pursuing Data Science and Artificial Intelligence degree
Interest in LLMs, AI agents, and modern AI systems
Passion for building intelligent applications
Basic Python programming knowledge recommended

16-Week Interactive Curriculum

Week 1

Transformer Architecture

Topics:

The Encoder-Decoder Framework
Introduction to transformer architectures and their evolution
The Hugging Face Ecosystem: Models, Datasets, Tokenizers, Pipelines
Practical: First experience with Hugging Face Pipelines

ILO: K1

Reference: Ref 1 - Chapter 1 + Ref 7

Project Milestone: Foundation Setup

Form teams (2-3 students) and create GitHub repository
Set up Python project structure, install dependencies (LangChain, LangGraph, ChromaDB)
Write a Transformer Analysis Document: explain the architecture of your chosen LLM backbone (attention, layers, tokenization)
Configure API keys (.env file for OpenAI/Anthropic/Tavily)

Deliverable: GitHub repo + Transformer Analysis (1-2 pages)

Start Learning

Week 2

Text Classification

Topics:

Text Classification with Transformers, Sentiment Analysis
Feature extraction vs fine-tuning approaches
Evaluation metrics: Accuracy, F1, Precision, Recall
Practical: Build a text classifier with fine-tuned transformer

ILO: K1, P1

Reference: Ref 1 - Chapter 2

Project Milestone: Classification Agent

Build a Classification Agent that categorizes research sources by type (paper, blog, news, documentation) and relevance (high/medium/low)
Use LLM-based classifier or fine-tuned transformer (BERT)
Implement evaluation metrics (accuracy, F1) on a test set of 20+ sources
Integrate into the project pipeline as a standalone module

Deliverable: Working classifier + evaluation report

Start Learning

Week 3

Named Entity Recognition (NER)

Topics:

Token Classification with BERT for NER
BIO/IOB Tagging scheme
Multilingual NER (Arabic and English)
Fine-tuning BERT for custom NER tasks
Evaluation per entity type

ILO: P1, T1

Reference: Ref 1 - Chapter 4

Project Milestone: NER Agent

Build an NER Agent that extracts entities from research text: people, organizations, technologies, dates, locations
Use SpaCy NER pipeline or LLM-based entity extraction
Output structured entity table with entity type, value, and frequency
Test on 5+ research articles and evaluate per-entity-type accuracy

Deliverable: Working NER module + entity extraction samples

Coming Soon

Week 4

LangChain Fundamentals

Topics:

Chains, Prompt Templates, Memory, Output Parsers
Document Loaders, Text Splitters, Vector Stores
Retrieval-Augmented Generation (RAG) with LangChain
Building QA systems over documents

ILO: K1, P1

Reference: Ref 4

Project Milestone: LangChain Components

Build the project's prompt templates for each agent (Research, Classifier, NER, Analyzer, Writer, Critic)
Implement document loaders and text splitters for web-scraped content
Set up ChromaDB vector store with embeddings for storing research findings
Build a basic RAG chain that retrieves and answers questions from stored sources

Deliverable: LangChain modules + working vector store

Start Learning Chat with Databases

Week 5

LangGraph

Topics:

Agent Orchestration as Graphs
Core concepts: Nodes, Edges, State, Conditional Routing
Building stateful, cyclical agent graphs
Checkpointing and Human-in-the-Loop
Tool integration within LangGraph nodes

ILO: I1, P2

Reference: Ref 4 - LangGraph Docs

Project Milestone: LangGraph Orchestrator

Build the LangGraph StateGraph skeleton with all agent nodes connected
Define the shared state schema (TypedDict with all fields)
Implement conditional routing: Orchestrator decides which agent runs next
Add state checkpointing (save/resume pipeline progress)
Integrate Classification + NER agents as graph nodes

Deliverable: Working graph with 3+ connected agents

Coming Soon

Week 6

Large Language Models (LLMs)

Topics:

Evolution from BERT/GPT-2 to modern LLMs
Model families: GPT-4, LLaMA, Gemini, Claude, Mistral, DeepSeek
Open vs Closed models, Prompt Engineering
Zero-shot, Few-shot, Chain-of-Thought
Running local models with Ollama

ILO: K1, K2

Reference: Ref 3 + Lecture Notes

Project Milestone: LLM Configuration & Benchmarking

Configure LLM parameters for each agent: model selection, temperature, max tokens, system prompts
Benchmark at least 2 different models (e.g., GPT-4 vs Claude vs Mixtral) on report quality
Implement token usage tracking and cost estimation per pipeline run
Optional: Set up Ollama for local model inference

Deliverable: LLM benchmark comparison table + cost analysis

Start Learning

Week 7

Mixture of Experts (MoE)

Topics:

Sparse Activation and MoE Architecture
Router/Gating Network, Expert Networks, Top-K Selection
Key models: Switch Transformer, Mixtral 8x7B, DeepSeek-MoE
Dense vs MoE comparison and trade-offs

ILO: K1, T1

Reference: Ref 5 + Scientific Papers

Project Milestone: MoE Analysis + Checkpoint 1 CHECKPOINT 1

Write an MoE Comparison Report: analyze MoE vs dense model architectures with benchmarks
Optional: Integrate Mixtral 8x7B as a cost-efficient model option in the pipeline
Implement model selection logic: use MoE for classification/NER, dense model for writing

CHECKPOINT 1: Demo NLP pipeline (Classification + NER + Research Agent) to instructor

Start Learning

Week 8

Midterm Exam

Comprehensive exam covering Weeks 1-7: Transformers, Text Classification, NER, LangChain, LangGraph, LLMs, and MoE

Week 9

AI Agents & ReAct

Topics:

What is an AI Agent (LLM + Tools + Memory + Planning)
The ReAct Framework: Reasoning and Acting (Yao et al., 2023)
The loop: Thought, Action, Observation
Tool Use: Web Search, Code Execution, APIs
Agent Memory: short-term and long-term

ILO: K2, P2

Reference: Ref 2 + Ref 4 + Ref 6

Project Milestone: Research Agent (ReAct)

Build the Research Agent with ReAct pattern: Thought -> Action -> Observation loop
Integrate Tavily/Brave Search API + web scraper as agent tools
Implement iterative search: agent decides when to search more or stop
Relevance scoring: LLM rates each source 1-10
Store findings in ChromaDB with source URLs for citation

Deliverable: Research Agent collects 10+ relevant sources per topic

Start Learning

Week 10

Multi-Agent Systems

Topics:

Agentic Design Patterns (Reflection, Tool Use, Planning, Multi-Agent)
Workflow Patterns: Sequential, Parallel, Hierarchical, Debate
Frameworks: LangGraph, CrewAI, AutoGen
Agent-to-Agent (A2A) protocols

ILO: K2, I1, P2

Reference: Ref 3 + Ref 4 + Ref 6

Project Milestone: Full Multi-Agent Pipeline

Integrate all agents into the LangGraph: Research -> Classification -> NER -> Analyzer -> Writer -> Critic
Build the Analyzer Agent: theme extraction, contradiction detection, outline generation
Build the Writer Agent: section-by-section writing with citations [1], [2]
Build the Critic Agent: scoring rubric (accuracy, completeness, clarity) + APPROVE/REVISE loop
Implement human-in-the-loop checkpoint before writing phase

Deliverable: End-to-end pipeline generates a complete research report

Start Learning A2A Protocol

Week 11

Model Context Protocol (MCP)

Topics:

The open standard connecting AI assistants to data and tools
MCP architecture: Hosts, Clients, Servers
Resources, Tools, Prompts, and Sampling
Building MCP servers and integrating with AI applications

ILO: K2, P2

Reference: Lecture Notes + MCP Docs

Project Milestone: MCP Integration CHECKPOINT 2

Expose agent tools as MCP servers (web search, scraper, NER, classifier)
Agents consume tools via MCP clients for standardized tool integration
Build at least 2 custom MCP tool servers (e.g., search server + NER server)
Implement error handling and retry logic across the full pipeline

CHECKPOINT 2: Demo full multi-agent pipeline with MCP to instructor + peer feedback

Start Learning

Week 12

Large Action Models (LAMs)

Topics:

Evolution: LLM to Agent to LAM
LAM vs Agent: natively trained on actions
Examples: Rabbit R1, Anthropic Computer Use, OpenAI Operator
Function Calling, Structured Outputs, GUI Understanding
Browser Automation, Safety and ethics

ILO: K2, P2, T1

Reference: Ref 6 + Lecture Notes

Project Milestone: Browser Agent (LAM)

Build a Browser Agent using Playwright/Selenium for web actions
Capabilities: navigate to pages, fill search forms, extract structured data from tables
Integrate as an optional research tool: when API search isn't enough, agent browses the web
Implement safety constraints: URL allowlist, action timeouts, no login/payment actions

Deliverable: Browser Agent extracts data from 3+ websites

Coming Soon

Week 13

Diffusion Models

Topics:

Generative AI and Diffusion Models fundamentals
Forward and reverse diffusion process
Denoising architectures (U-Net, Transformers)
Applications in image generation (Stable Diffusion, DALL-E)
Practical generation and fine-tuning

ILO: K1, P1

Reference: Ref 8 + Lecture Notes

Project Milestone: Illustration Agent + UI

Build an Illustration Agent using Stable Diffusion API or DALL-E to generate figures for reports
Auto-generate: concept diagrams, comparison charts, architecture visualizations
Embed generated images in the Markdown/PDF report output
Build Streamlit UI: topic input, real-time progress, agent reasoning panels, PDF download

Deliverable: Reports with AI-generated illustrations + working web UI

Start Learning

Week 14

Project Development & Final Testing

Dedicated week for final development, bug fixes, and instructor consultation

Project Milestone: Competition Preparation

End-to-end testing on 5 diverse research topics
Performance benchmarking: speed, token cost, report quality
Write README and technical documentation
Record 5-minute demo video of the full pipeline
Prepare 10-minute presentation slides
Submit all deliverables to GitHub before competition day

Deliverable: Final GitHub repo + docs + demo video + slides

Week 15

Competition Day & Presentations

Live Competition: All teams receive 3 unseen research topics and run their systems in real-time (30 min/topic)

Competition Day Protocol

3 unseen topics revealed at the start of the session
All teams run systems simultaneously -- 30-minute limit per topic
Reports anonymized and scored by judging panel (instructor + peers)
10-minute live presentation per team: architecture, demo, Q&A
Each team reviews 2 other teams' reports (peer review graded)
Awards: Gold, Silver, Bronze + Best Innovation + Best Report + Fastest System

Competition scores = 40% of project grade

Week 16

Final Exam

Comprehensive final exam covering all course material with emphasis on Agents, Multi-Agent Systems, MCP, and LAMs

Test Your Knowledge

1. What is the core mechanism that allows transformers to process sequences effectively?

A) Recurrence

B) Self-Attention

C) Convolution

D) Pooling

2. In the ReAct framework, what does the cycle consist of?

A) Input, Process, Output

B) Thought, Action, Observation

C) Encode, Decode, Generate

D) Query, Key, Value

3. What is the main advantage of Mixture of Experts (MoE) models?

A) Smaller model size

B) Better training data

C) Sparse activation reduces compute while maintaining performance

D) Simpler architecture

Project Portfolio

Build hands-on projects throughout the course that demonstrate mastery of modern AI systems

Arabic QA System

Build a question-answering system using RAG and LangChain over Arabic documents

RAG LangChain Week 6

Research Assistant

Multi-agent research assistant that searches, summarizes, and synthesizes information

Multi-Agent LangGraph Week 10

Smart Tutor

AI-powered tutoring agent that adapts to student needs and provides personalized feedback

Agents LLMs Week 9

Document Analyzer

Intelligent document analysis system with RAG and vector database integration

RAG Vector DB Week 4

Code Review Agent

ReAct-based agent that reviews code, suggests improvements, and explains issues

ReAct LangGraph Week 9

Browser Automation Agent

Large Action Model that automates web tasks through browser interaction

LAM Playwright Week 12

Assessment & Grading

Component	Weight	Description
Assignments	10%	Weekly assignments and exercises throughout the course
Rubric (Project)	20%	Comprehensive course project with rubric-based evaluation
Mid Exam	30%	Midterm exam covering Weeks 1-7
Final Exam	40%	Comprehensive final exam covering all course material

Exam Details

Midterm Exam (Week 8):

Covers Weeks 1-7
Transformers, Text Classification, NER
LangChain, LangGraph
LLMs, Mixture of Experts

Final Exam (Week 16):

All course material
Emphasis on AI Agents, ReAct
Multi-Agent Systems, MCP
Large Action Models, Diffusion Models

Weekly Schedule

In-Campus Session: 3 hours

Theory + Live demonstrations + Hands-on practice + Interactive lectures with real-time coding

Self-Study: 4-6 hours per week for project completion, practice, and experimentation

Total weekly commitment: 7-9 hours (3 hours in-campus + 4-6 hours independent work)

Office Hours

Every Day: 10:00 AM - 12:00 PM

Innovation Center

Available for: Technical questions, project guidance, debugging support, and consultation

Drop-in welcome or schedule an appointment via email

Resources & Support

Required Software

Python 3.12+
PyTorch
Hugging Face Transformers
LangChain & LangGraph
Ollama
VS Code

Hardware

Minimum: 8GB RAM, i5 processor, 50GB storage, webcam

Recommended: 16GB RAM, i7/Ryzen 7, dedicated GPU, 100GB SSD

References & Reading

Ref 1

NLP with Transformers - Tunstall, von Werra, Wolf (O'Reilly, 2022)

Ref 2

"ReAct: Synergizing Reasoning and Acting" - Yao et al., 2023

Ref 3

AI Engineering - Chip Huyen (O'Reilly, 2025)

Ref 4

LangChain & LangGraph Documentation - Official Docs

Ref 5

"Mixture of Experts Explained" - Hugging Face Blog

Ref 6

A Visual Guide to LLM Agents - Maarten Grootendorst, 2025

Ref 7

"Attention Is All You Need" - Vaswani et al., 2017

Ref 8

"Understanding Diffusion Models" - Calvin Luo, 2022

Computing Systems for Data Science and Artificial Intelligence

Course Overview

What You'll Learn

What You'll Build

Technologies

Who Should Take This Course?

16-Week Interactive Curriculum

Transformer Architecture

ILO: K1

Project Milestone: Foundation Setup

Text Classification

ILO: K1, P1

Project Milestone: Classification Agent

Named Entity Recognition (NER)

ILO: P1, T1

Project Milestone: NER Agent

LangChain Fundamentals

ILO: K1, P1

Project Milestone: LangChain Components

LangGraph

ILO: I1, P2

Project Milestone: LangGraph Orchestrator

Large Language Models (LLMs)

ILO: K1, K2

Project Milestone: LLM Configuration & Benchmarking

Mixture of Experts (MoE)

ILO: K1, T1

Project Milestone: MoE Analysis + Checkpoint 1 CHECKPOINT 1

Midterm Exam

AI Agents & ReAct

ILO: K2, P2

Project Milestone: Research Agent (ReAct)

Multi-Agent Systems

ILO: K2, I1, P2

Project Milestone: Full Multi-Agent Pipeline

Model Context Protocol (MCP)

ILO: K2, P2

Project Milestone: MCP Integration CHECKPOINT 2

Large Action Models (LAMs)

ILO: K2, P2, T1

Project Milestone: Browser Agent (LAM)

Diffusion Models

ILO: K1, P1

Project Milestone: Illustration Agent + UI

Project Development & Final Testing

Project Milestone: Competition Preparation

Competition Day & Presentations

Competition Day Protocol

Final Exam

Test Your Knowledge

Project Portfolio

Arabic QA System

Research Assistant

Smart Tutor

Document Analyzer

Code Review Agent

Browser Automation Agent

Assessment & Grading

Exam Details

Weekly Schedule

Office Hours

Resources & Support

Required Software

Hardware

References & Reading

Ref 1

Ref 2

Ref 3

Ref 4

Ref 5

Ref 6

Ref 7

Ref 8

Navigation