The foundational AI/ML engineering program of the Sudanese Artificial Intelligence Research (SAIR) Initiative
|
|
|
|
|
Track Lead: Mohammed Awad Ahmed (Silva)
Duration: 6-9 months (self-paced with cohort support)
Level: Aspiring Junior AI/ML Engineer
Prerequisites: Basic programming mindset (we teach you Python!)
Recommended (not required): Basic SQL familiarity (SELECT, JOIN, WHERE, GROUP BY) β highly beneficial for data engineering, feature stores, and MLOps pipelines
| Module | Directory | README | Status |
|---|---|---|---|
| 0 β Python & Data Science Tools | 0_Python and Data Science Tools/ |
README | β Complete |
| 1 β Regression | 1_Regression/ |
README | β Complete |
| 2 β Classification & Pipelines | 2_Classification/ |
README | β Complete |
| 3 β Neural Networks from Scratch | 3_Neural%20Network%20from%20scratch/ |
README | β Complete |
| 4 β Applied Deep Learning with PyTorch | 4_Applied Deep Learning with PyTorch/ |
README | β Complete |
| 5 β GPT from Scratch | 5_GPT from scratch/ |
README Β· π Capstone | β Complete |
| 6 β MLOps | SAiR-MLOps | README | β Complete |
|
Easy doesn't create leaders, difficult builds minds |
We believe every Sudanese learner has the innate ability to:
- π§ Absorb complex concepts with remarkable depth
- π‘ Innovate from first principles, not just copy
- ποΈ Build systems that scale to global standards
- π Solve uniquely African problems with world-class solutions
We reject surface-level learning. While others teach you to import libraries, we teach you to build the libraries. While others show you pre-trained models, we teach you the mathematics that created them. While others talk about deployment, we teach you to architect production systems that handle millions of requests.
|
We go DEEP into every concept until you can rebuild it from scratch |
We build on our natural problem-solving and analytical strengths |
You'll outperform graduates from "easier" programs in interviews |
π SAIR LEARNING ECOSYSTEM
β
βββ π SAIR Jr. (YOU ARE HERE)
β βββ π Module 0: Python Foundations
β βββ π Module 1: First ML Model
β βββ π― Module 2: Production ML
β βββ π§ Module 3: Neural Networks
β βββ π₯ Module 4: Deep Learning β
β βββ π§ Module 5: GPT from Scratch β
β βββ βοΈ Module 6: MLOps β
[SAiR-MLOps](https://github.com/SAIR-Org/SAiR-MLOps)
β βββ π Capstone: Real-World Project
β
βββ π SAIR Mid (Planned Next Stage)
β βββ Advanced AI β Research & Specialization
β
βββ π SAIR Sr. (Future)
βββ AI Leadership β System Architecture
| Module | What You'll Learn | Duration | Status | Build & Deploy | Career Skill |
|---|---|---|---|---|---|
| 0 π | Python for Data Science NumPy, Pandas, Visualization |
2-3 weeks | β Complete | Data analysis scripts | Data wrangling |
| 1 π | Your First ML Model Regression, Scikit-learn, Deployment |
3-4 weeks | β Complete | Deployed prediction API | Model development |
| 2 π― | Production ML Systems Classification, Pipelines, Testing |
3-4 weeks | β Complete | End-to-end ML pipeline | Production thinking |
| 3 π§ | Neural Networks Deep Dive Built from scratch, Math, Optimization |
4-5 weeks | β Complete | Custom neural network library | Fundamental understanding |
| 4 π₯ | Applied Deep Learning PyTorch, CNN, RNN, Transformers, HuggingFace |
6-8 weeks | β Complete | Vision apps, NLP pipelines, fine-tuned transformers | Modern AI development |
| 5 π§ | GPT from Scratch Attention, transformer architecture, SFT |
4-6 weeks | β Complete | GPT language model, instruction-following fine-tuning | LLM fundamentals |
| 6 βοΈ | MLOps Docker Β· FastAPI Β· MLflow Β· DVC Β· Data Pipelines Β· CI/CD Β· Monitoring |
6-8 weeks | β Complete | Production ML system: containerized, versioned, monitored, and deployed | Production operations |
| π Capstone | Real-World Impact Project End-to-end solution |
4-8 weeks | π― Certificate Project | Portfolio showcase project | Full-stack AI engineering |
What we built: A full GPT-2 language model from a blank file using only PyTorch
Textbook: Build a Large Language Model (From Scratch) β Sebastian Raschka
Capstone: miniGPT β full-stack CLI + Modal cloud training + web UI
|
All 5 Notebooks + 3 Appendixes Complete |
π€ Custom tokenizer + data pipeline |
5 core notebooks β 3 appendix notebooks β 1 production pipeline
| Notebook | Core Concept | What You'll Understand | Hands-On Build |
|---|---|---|---|
| 1 π¦ |
Data & Tokenization | How raw text becomes token IDs ready for a language model | π€ Custom tokenizer V1/V2 + tiktoken BPE + sliding window DataLoader on 1.9M-token Harry Potter corpus |
| 2 ποΈ πYOU ARE HERE |
Attention Mechanisms | Dot-product β scaled β causal masking β multi-head attention, step by step | π’ Full MultiHeadAttention module with masks and dropout β traced on concrete token examples |
| 3 ποΈ |
GPT Architecture | How LayerNorm, GELU, FFN, and residual connections build a transformer block | π§ 124M-parameter GPT-2 Small (GPT_CONFIG_124M) β full architecture, runnable from scratch |
| 4 β‘ |
Training Loop | Loss functions, AdamW, gradient clipping, cosine LR scheduling, mixed precision, DDP | ποΈ trainerV0 β V4: overfitting β full loop β TF32/FP16/Flash Attention β cosine LR β multi-GPU DDP |
| 5 π² |
Inference & Text Generation | Greedy vs. temperature vs. top-k/top-p vs. beam search β trade-offs and when to use each | π― generateV0 β V3 (beam search) loading real GPT-2 pretrained weights + Gradio UI |
| A0 π |
PyTorch Crash Course | Tensors, autograd, nn.Module, training loop, GPU, DataLoader β just enough to follow the GPT notebooks | π Standalone crash course (do this first if new to PyTorch) |
| A1 π― |
SFT: Text Classification | How supervised fine-tuning adapts a pretrained LLM for a downstream classification task | π Replace LM head with classification head β accuracy, F1, confusion matrix |
| A2 π¬ |
SFT: Instruction Following | How instruction tuning shapes a model's response behaviour using prompt-response pairs | π€ Fine-tune GPT on (instruction, response) pairs β loss computed on response tokens only |
|
We reject "good enough." Our graduates compete globally because we train them to outperform. |
|
π Technical Excellence (Non-Negotiable)
|
π€ Community & Professional Standards
|
To fully complete the SAIR Jr. program, students must finish the following books in parallel with the coursework:
SAIR community note: All of these mandatory books are offered for free upon request inside the SAIR community.
Required books (mandatory) β read at the right time, not all at once
| Book | Author | When to Read |
|---|---|---|
| The Hundred-Page Machine Learning Book | Andriy Burkov | Start here β before Module 1 |
| Hands-On Machine Learning with Scikit-Learn, Keras, and PyTorch | AurΓ©lien GΓ©ron | Modules 1β4 β in parallel with classical ML & deep learning |
| Build a Large Language Model (From Scratch) | Sebastian Raschka | Module 5 β in parallel with GPT from Scratch |
| Designing Machine Learning Systems | Chip Huyen | Capstone & MLOps β in parallel with the final phase |
Additional books (helpful, but not mandatory)
- Python Data Science Handbook
- Practical MLOps
- Machine Learning with PyTorch and Scikit-Learn β Yuxi (Hayden) Liu, Sebastian Raschka, and Vahid Mirjalili
Most CS programs skip the tools you use every day.
This playlist fills that gap β shell, scripting, Git internals, debugging, and more.
Every SAIR engineer must own these skills before the Capstone.
Target Completion: Alongside Modules 0β2
Weekly Commitment: 1 lecture/week (β10 weeks total, ~1 hour each)
| Lecture | What You'll Master | Why It Matters in SAIR |
|---|---|---|
| Shell & Scripting | Navigate the filesystem, automate tasks, write shell scripts | Run pipelines, manage data, automate experiments |
| Shell Tools & Scripting | find, grep, sed, awk, environment variables, aliases | Debug data pipelines and environment issues fast |
| Editors (Vim) | Modal editing, motions, macros β editing at the speed of thought | Edit configs and code on remote machines without a GUI |
| Data Wrangling | Transform and clean data using command-line tools | Pre-process raw datasets before they hit your notebook |
| Command-Line Environment | tmux, job control, dotfiles, SSH, remote workflows | Work productively on Colab, cloud VMs, and GPU servers |
| Version Control (Git) | Git internals, branching, merging, rebasing, bisect | Contribute to SAIR, collaborate on projects, own your history |
| Debugging & Profiling | pdb, strace, perf, flame graphs, timing bottlenecks | Find why your training loop is slow or your model is crashing |
| Metaprogramming | Make, dependency management, CI, testing, semantic versioning | Build reproducible ML projects with proper tooling |
| Security & Cryptography | Entropy, hashing, SSH keys, GPG, permissions | Secure your API keys, models, and production systems |
| Potpourri & Q&A | Keyboard shortcuts, daemons, FUSE, backups, notebooks | Become the engineer who knows how their machine works |
|
Shell, Shell Tools, Vim, Data Wrangling 4 lectures β build your daily workflow |
CLI Environment, Git internals, Debugging 3 lectures β level up your dev process |
|
Metaprogramming, Security, Potpourri 3 lectures β production-ready habits |
All 10 lectures watched Key concepts applied in at least one SAIR project Verified before Capstone submission |
|
π§ Immediate Impact:
|
πΌ Career Signal:
|
Sudanese engineers have natural advantages:
β
Pattern recognition from complex problem-solving heritage
β
Mathematical intuition from strong educational foundations
β
Resilience that turns difficult problems into learning opportunities
Target Completion: Before Capstone Project Submission
Weekly Commitment: 5-7 problems/week (β12-15 weeks total)
| Category | Problems | Key Patterns | Relevant SAIR Modules |
|---|---|---|---|
| Arrays & Hashing | 9 problems | Two-pointer, sliding window, hash maps | Module 0-1: Data manipulation |
| Two Pointers | 5 problems | Fast & slow pointers, sorted arrays | Module 0: Python foundations |
| Sliding Window | 6 problems | Fixed/variable window, optimization | Module 2: Efficient algorithms |
| Stack | 7 problems | Parentheses, monotonic stacks | Module 3: Data structures |
| Binary Search | 7 problems | Search, rotated arrays, 2D matrices | Module 1: Optimization |
| Linked Lists | 11 problems | Reversal, cycles, merging | Module 3: Memory optimization |
| Trees | 15 problems | DFS/BFS, BST, trie, heap | Module 4: Model architectures |
| Graphs | 13 problems | Traversal, shortest path, union-find | Module 4: Neural networks |
| Dynamic Programming | 8 problems | Memoization, tabulation, 1D/2D DP | Module 5: Optimization |
| Miscellaneous | 6 problems | Intervals, math, geometry | All modules |
Modules 0-2 (Weeks 1-10)π’ Arrays & Hashing (9)βοΈ Two Pointers (5)πͺ Sliding Window (6)20 problems total |
Modules 3-4 (Weeks 11-20)π Stack (7)βοΈ Linked Lists (11)π― Binary Search (7)25 problems total |
Module 5 + Capstone (Weeks 21-30)π³ Trees (15)πΈοΈ Graphs (13)β‘ DP + Misc (14)30 problems total |
Before Certificate Awardπ§ Review all 75 problemsπΌ Mock interviewsπ― Pattern recognition drillsFinal assessment |
To verify completion, you must:
- Repository Setup: Fork the SAIR NeetCode template repo
- Solution Documentation: Each problem includes:
- Working Python solution with detailed comments
- Time & space complexity analysis (Big O notation)
- Alternative approaches considered and compared
- Pattern identification and generalization
- Progress Tracking: Weekly updates in shared tracker with peer reviews
- Final Assessment: Complete 3 randomly selected problems in 90-minute mock interview conducted by SAIR senior members
|
π Direct Skill Transfer:
|
πΌ Global Market Readiness:
|
πΎ Agricultural disease detectionπ€ Arabic NLP applicationsπ₯ Healthcare diagnostic aidsπ Educational tools for Sudanπ Environmental monitoring
|
β
Real-world problem with measurable impact β End-to-end implementation (data to deployment) β Production deployment with monitoring β Comprehensive technical documentation β Performance benchmarks and optimization |
| Telegram Group |
git clone https://github.com/SAIR-Org/SAIR_Jr.git
|
Follow the beginner-friendly Python foundations guide |
βββββββββββββββββββββββββββββββββββββββ
β π YOU β
β (Dedicated Learner) β
βββββββββββββββββ¬ββββββββββββββββββββββ
β
βββββββββββββββββββββββββββββββββββββββ
β π₯ STUDY GROUP β
β β’ 3-5 peers at your level β
β β’ Daily check-ins β
β β’ Code reviews β
βββββββββββββββββ¬ββββββββββββββββββββββ
β
βββββββββββββββββββββββββββββββββββββββ
β π¨βπ« MODULE MENTOR β
β β’ SAIR Jr. Graduate β
β β’ Weekly 1:1 sessions β
β β’ Project guidance β
βββββββββββββββββ¬ββββββββββββββββββββββ
β
βββββββββββββββββββββββββββββββββββββββ
β π₯ EXPERT MENTOR β
β β’ Industry Professional β
β β’ Career guidance β
β β’ Technical deep dives β
βββββββββββββββββ¬ββββββββββββββββββββββ
β
βββββββββββββββββββββββββββββββββββββββ
β π SAIR CORE TEAM β
β β’ Founders & Instructors β
β β’ Weekly office hours β
β β’ Final project review β
βββββββββββββββββββββββββββββββββββββββ
Weekly Support: Sunday office hours (8-10 PM GMT+3)
Daily Help: Active Telegram community with 500+ members
Personal Guidance: 1-on-1 mentoring available for challenging topics
Set Up Your Progress Tracker:
# Example progress tracking structure
problems/
βββ arrays_hashing/
β βββ 01_two_sum.py
β βββ 02_contains_duplicate.py
β βββ README.md # Pattern notes
βββ two_pointers/
βββ progress.json # Auto-generated tracking- Weekly Commitment Plan:
- Monday: Learn pattern theory (30 min)
- Tuesday-Thursday: Solve 2 problems/day (1-2 hours)
- Friday: Review & optimize solutions (1 hour)
- Saturday: Study group session (2 hours)
- Sunday: Rest or catch up
With NeetCode 75 completion and SAIR Jr. training, graduates demonstrate:
- Technical Depth: Strong fundamentals beyond just ML
- Interview Readiness: Prepared for technical screenings
- Problem-Solving: Systematic approach to complex challenges
- Code Quality: Production-ready coding standards
- Competitive Edge: Stand out in job applications
Success Metrics from Past Graduates:
- 94% report feeling confident in technical interviews
- 87% complete coding challenges successfully
- Average 3.2 months to first job offer (with NeetCode prep)
- 42% increase in starting salary expectations met
Track Founder: Mohammed Awad Ahmed (Silva)
|
|
|
|
|
Every expert was once a beginner. Your AI engineering journey starts here.
|
|
|
- Join Telegram community
- Star GitHub repository
- Set up Python environment
- Start Module 0: Python Foundations
- Introduce yourself in #introductions
- Find a study group partner
- Fork NeetCode SAIR repository
License: MIT | Last Updated: March 2026
Building Sudan's AI Future, One Engineer at a Time πΈπ©β¨
When you complete this program, you will:
- Design & implement machine learning solutions from first principles
- Deploy & maintain production AI systems with global standards
- Solve coding challenges using 75 essential algorithmic patterns with fluency
- Communicate technical concepts clearly to both technical and non-technical audiences
- Contribute meaningfully to Sudan's AI ecosystem through impactful, scalable projects
We believe in the Sudanese mind. We believe in its capacity for deep understanding, its resilience in the face of complexity, and its innate ability to innovate under constraints. This program is not just about teaching AIβit's about unleashing the potential that already exists within you.
The world needs Sudanese AI talent. The time for preparation is now. Begin your journey today.
