📊 Data to Decisions — Interactive Data-Literacy Workshop Platform

A multi-page Streamlit app that teaches non-programmers the full data lifecycle as live, hands-on demos — one URL, slides and demos in the same sidebar. It is the interactive backbone of the From Data to Decisions workshop by Neurobotics Academy (Eng. Mulham Fetna).

Every demo follows one small fictional business — Nour Store — from raw, messy data all the way to a decision, so the audience watches a single story become a decision.

▶ Try the live lab: https://mulhamfetna-datalab.streamlit.app/ 🔗 Website: https://datalab.mulhamfetna.com

Run

./run.sh

or

pip install -r requirements.txt
streamlit run Home.py

Opens at http://localhost:8501. Override the port with PORT=9000 ./run.sh.

What's inside

Slides (native pages): Why Data · The Lifecycle · Learning Roadmap · Tech Roles · Career · Close

Collect: Web Scraper · REST API · Synthetic Data · Public Datasets · Internal Data · Qualitative→Quantitative · Survey

Transform (on the shared Nour Store data): Clean · Filter · Analyze · Engineer (star schema → SQLite) · Predict (a scikit-learn classifier) · Report

Data Lab (35 modules across 9 lifecycle stages):

• Acquire — streaming, file formats, SQL extraction, data integration
• Label — interactive labeling, active learning, weak supervision
• Clean — profiling, deduplication, imputation, outlier treatment
• Encode — tokenization, quantization, embeddings, categorical encoding, scaling
• Mine — market-basket, clustering/segmentation, anomaly detection
• Model — forecasting, recommendation, A/B testing, sentiment, image classification
• Scale — big-data chunking, compression, indexing, warehouse-vs-lake
• Govern — PII masking, fairness audit, data drift, lineage
• Serve — model-as-API, monitoring/retraining, orchestration

Second wave — judgment, GenAI, and teaching layers:

• Data Judgment — Simpson's paradox, correlation vs causation, sampling bias, misleading charts, confidence intervals & p-hacking
• GenAI / LLM — prompt playground, RAG, summarization, structured extraction, zero-shot classifier, guardrails & hallucination. Each has a free-provider live path (Groq / OpenRouter / Hugging Face / local Ollama, via their OpenAI-compatible endpoints) and an honest offline simulation, so it runs on stage with no key.
• Explainability (XAI) — global feature attribution, single-prediction explanation
• Advanced Analytics — price optimization, Monte Carlo, geospatial, network analysis, matrix-factorization recommender
• Teaching Layers — capstone path, role-based tracks, per-epic quizzes, bilingual glossary, case-study cards

Every module was built on its own branch and merged via pull request; see the repository's closed Issues and Releases for the full history.

Architecture

• workshop/ — pure, UI-free, unit-tested logic (store_data, cleaning, queries, pipeline, model, synth, sources, report, visuals, live_api); the one exception is ui.py, the shared Streamlit header helper.
• pages/ — thin Streamlit views over that logic (Streamlit's multipage convention).
• contact_scraper/ — the bundled contact-scraper package powering the Scraper demo.
• assets/ — bundled datasets and preview images (self-contained, offline-safe).
• tests/ — pytest suite for the logic modules plus AppTest smokes for every page.

Test

python3 -m pytest -m "not live"    # offline suite
python3 -m pytest -m live          # network smoke tests

Honest limitations

• Visuals are generated, not photographed — matplotlib figures, graphviz diagrams, and inline SVG — so the app runs fully offline; it does not embed external stock images.
• The Predict demo trains a small, real model on synthetic data; its "high-value" label is spend-derived, so accuracy is high partly by construction — it illustrates the idea, it is not a hard predictive benchmark.
• The Synthetic demo does real statistical generation + validation; it does not train a GAN/transformer live, and says so — it shows the principle those methods scale up.
• The Scraper demo's Syrian (.sy) targets are geo-restricted and only reachable from within Syria; scraping runs server-side, so run the app locally on a Syrian connection for those.
• The GenAI demos call a real model only when a free provider key is set (GROQ_API_KEY, OPENROUTER_API_KEY, HF_TOKEN, or a local OLLAMA_HOST); with none set they run a clearly-labeled offline simulation and never present it as a real model.

Citation

If you use this platform, please cite it. Archived on Zenodo — concept DOI 10.5281/zenodo.21427808 (always resolves to the latest version). GitHub also renders a "Cite this repository" button from CITATION.cff.

@software{fetna_datalab_2026,
  author    = {Fetna, Mulham},
  title     = {Data to Decisions: An Interactive Data-Literacy Workshop Platform},
  year      = {2026},
  publisher = {Zenodo},
  doi       = {10.5281/zenodo.21427808},
  url       = {https://github.com/mulhamfetna/Data-Lab}
}

License

AGPL-3.0-or-later © 2026 Mulham Fetna / Neurobotics Academy.