Class of ML · Cohort 1
Active Cohort · Wrapping UpFrom ML Fundamentals
to Shipping Deployed AI.
An 8-week intensive engineering cohort (70–90 min sessions/week) that ran from August 1, 2026, and is now wrapping up. Guided by senior mentors and 2 dedicated teaching assistants per mentor, every student has built and shipped a live, deployed, portfolio-ready ML/AI system.
70–90 min weekly deep-dives with hands-on live code reviews
2 dedicated teaching assistants per mentor for real-time debug support
Every single student ships a live, portfolio-ready ML/AI product
The 8-Week Journey
What Students Built & Mastered
Over 8 weeks, students progressed step-by-step from raw problem formulation and evaluation rigor to deep learning, LLM architectures, RAG pipelines, and full production deployment.
Thinking Like an ML Engineer: Problem Framing & Honest Evaluation
Students learned to frame a real problem as an ML task tied to a business outcome, choose and defend the right evaluation metric (Precision, Recall, F1, ROC-AUC), spot data leakage, and set up an honest train/val/test split.
Core Competencies
Working With Real Data: Feature Engineering & Data Storytelling
Cleaning messy real-world data, engineering features that measurably improve a model, handling class imbalance, and communicating a data insight in one slide.
Core Competencies
Winning With Tabular Data: Decision Trees to XGBoost
Decision trees to random forests to gradient boosting, building and tuning XGBoost/LightGBM, reasoning about bias-variance, producing a well-evaluated capstone baseline.
Core Competencies
Neural Networks Demystified: How Deep Learning Actually Works
Neurons, layers, activations, backprop and gradient descent intuition, diagnosing overfitting, building a neural network from scratch.
Core Competencies
From Transformers to LLMs: The Engine Behind Modern AI
Attention and why Transformers took over, tokenization, pretraining vs fine-tuning, why LLMs hallucinate, prompt engineering as a disciplined practice.
Core Competencies
Building With LLMs: Retrieval-Augmented Generation (RAG)
Embeddings, vector search, RAG architecture, building a citation-grounded RAG pipeline, evaluating for hallucination.
Core Competencies
Shipping AI: Deployment & Production Basics
Turning the capstone into a clickable demo, exposing an API endpoint, making it reproducible, deploying to a public URL.
Core Competencies
Launch: Portfolio, Positioning & Landing the Role (Final Presentations)
Writing a README that sells the project, positioning LinkedIn/resume for a target role, final walkthroughs, junior AI/ML interview prep.
Core Competencies
Student Output
Shipped Capstone Projects
Every student concludes the cohort with a live, reproducible AI application backed by public API endpoints, clean GitHub repositories, and interactive user interfaces.
Featured Capstone Architecture
Production ML Pipeline & Public Inference API
[INFO] Loading held-out benchmark dataset (n=12,480 samples)...
[INFO] Model Architecture: GBDT + Dense Neural Residual Fusion
[METRIC] Out-of-Fold F1-Score: 0.924 | ROC-AUC: 0.967
[METRIC] Inference P95 Latency: 18.4ms (Batch Size = 1)
[SUCCESS] Verified zero data leakage across temporal splits.
[DEPLOY] Container image pushed: registry.afrium.dev/models/capstone-v1:liveProduction Engineering Pipeline
Lahore AQI Forecast & Air Quality Early Warning System
Trained gradient-boosted ensembles on 4 years of Pakistani air quality and meteorological sensor streams, deploying a real-time 48-hour particulate matter forecast API.
QanoonAI: Pakistani Legal RAG & Statute Citation Search
Vectorized Pakistan Supreme Court precedents and the Penal Code with hybrid semantic-lexical search to synthesize citation-backed legal briefs without hallucination.
CropGuard: Agricultural Disease Diagnostic Pipeline
Lightweight convolutional network fine-tuned on Pakistani wheat and cotton crop leaf pathology, packaged into a mobile-friendly inference endpoint.
By the final week, every student has a
deployed, portfolio-ready AI project
and a story to tell about it.
Students graduate with live interactive demos, clean system architecture READMEs, and the technical depth required to clear rigorous AI/ML engineering interviews.