Implement a full Transformer encoder — multi-head attention, positional encoding, layer norm — in PyTorch from scratch. Train on a classification task. No HuggingFace shortcuts.
Implement scaled dot-product attention and multi-head attention from scratch in PyTorch
Build sinusoidal positional encoding and understand why position matters in Transformers
Assemble a complete Transformer encoder block with residual connections and layer norm
Train an encoder classifier end-to-end on a real text classification dataset
Scrape content, clean it, auto-generate instruction-response pairs using an LLM, score quality with an evaluator model, and output a production-ready JSONL dataset.
Build an async web scraping pipeline using httpx and BeautifulSoup
Clean, deduplicate, and validate raw text content at scale
Auto-generate instruction-response training pairs using an LLM
Score dataset quality using an LLM judge and apply rule-based filters
Build a lightweight feature store that computes, caches, and serves ML features. Connect it to both a training pipeline and a real-time prediction API.
Understand the purpose of a feature store and the training/serving skew problem
Compute, version, and store ML features in Redis (real-time) and PostgreSQL (historical)
Connect a feature store to both a training pipeline and a live inference API
Verify consistency between features used in training and features used in serving
Train an LSTM on real product reviews. Run the same data through a zero-shot LLM classifier. Compare accuracy, latency, and cost — understand where each belongs.
Build, train, and evaluate an LSTM text classifier in PyTorch
Implement a zero-shot LLM classifier and measure its performance
Compare trained models vs LLMs on accuracy, latency, and cost per prediction
Understand the trade-off space: when to use fine-tuned models vs zero-shot LLMs
Generate a benchmark dataset by prompting an LLM across many scenarios, score outputs on multiple criteria, and produce a structured eval report with failure analysis.
Design a multi-dimensional scoring rubric for evaluating LLM outputs
Generate structured evaluation datasets using async LLM API calls
Build an LLM-as-judge pipeline that scores model responses automatically
Calculate inter-rater agreement between human and automated scoring
Forecast product sales using Prophet and ARIMA. Build a Streamlit dashboard that visualizes predictions vs. actuals and explains seasonality and trend components.
Identify trend, seasonality, and stationarity in a real time series dataset
Train and evaluate both Prophet and ARIMA forecasting models
Compare model performance using MAE, RMSE, and MAPE
Visualize forecast results with confidence intervals in Streamlit
Build a content-based recommendation system using TF-IDF and cosine similarity. Enter a movie, get 5 similar recommendations. Intuition-builder for how embeddings work
Understand TF-IDF vectorization and cosine similarity intuitively
Build a content-based recommendation engine from scratch
Implement a nearest-neighbour search using cosine similarity matrices
Connect recommendation logic to an interactive Streamlit UI\
Train a linear regression model to predict student exam scores from study hours, attendance, and test scores. Deploy it as a simple Streamlit app with a prediction form.
Build and train a linear regression model using scikit-learn
Encode categorical variables and prepare features for ML models
Evaluate regression models using RMSE and R² metrics
Save a trained model to disk and load it for inference
Take a real-world dataset, clean it, explore distributions, correlations, and outliers, and visualize everything in a Streamlit dashboard. The foundation of all data work
Load and clean a real-world dataset using pandas
Identify and handle missing values, duplicates, and outliers
Visualize distributions and correlations using matplotlib and seaborn
Build an interactive multi-chart Streamlit dashboard
Every challenge includes detailed documentation, technical constraints, and automated evaluation scripts to ensure you have everything you need to succeed.