zaki_rasa.ipynb
Python 3 · idle
Portrait of Zaki Rasa

Zaki Rasa

> Data Scientist & Founder

forecasting/full-stack products/ML pipelines/data governance

I build forecasting systems end to end: from deep-learning research, to production ML, to a product people pay for.

// Porto, Portugal

About Zaki Rasa

In [1]:
zaki.about()
Out[1]:

I build forecasting systems from first experiment to production. At SISQUAL WFM in Porto I take workforce-forecasting models from research into production APIs. On my own, I built ForecastMyShop, a forecasting service for independent retailers, from the model all the way to billing.

I care about clean data, honest back-testing, and the simplest model that actually wins. Lately I also work on making enterprise data AI-ready: data catalogs, governed semantic layers, and RAG- and MCP-based assistants.

Impact in numbers

In [2]:
zaki.impact()
Out[2]:
resultwhatwhere
0−18%Forecast error (WMAPE), via feature engineering and Optuna tuningSISQUAL WFM
1+9%Accuracy vs. an autoregressive baseline, on 20M rows from 119 pharmaciesINESC TEC research
2−40%Manual querying and reporting time for stakeholder teamsSISQUAL WFM
39Interface languages in a product built soloForecastMyShop
4500+University IT staff using the national HEMIS platformMinistry of Higher Education

5 rows × 3 columns

In [3]:
zaki.projects["forecastmyshop"].show()
Out[3]:

ForecastMyShop 2026 – present

A forecasting service for independent shops: pharmacies, grocery, fashion, beauty, restaurants and online stores. Owners upload the sales file they already have and get per-product forecasts with ranges, order suggestions and insights.

Visit forecastmyshop.com →
Example product forecast Twenty weeks of sales trending from about 42 to 70 units, followed by an eight-week forecast rising to about 83 units, with a range that widens further out. 406080 forecast → week 1today+8 wk
// illustrative data, not a real shop
  • Engine built around Chronos-2, Amazon's open-source foundation model, alongside classical methods. Each forecast is back-tested on the shop's own sales; the best method wins, and a simple baseline stays when it matches.
  • A data-quality layer for messy real inputs (Excel, till exports, CSV, copied tables) that catches gaps, spikes, bad dates and duplicates before any forecast runs.
  • The full stack, solo: accounts and plans, credit-based billing, Excel/PDF/CSV exports, a customer API, and a 9-language interface.
  • Privacy by design: EU-hosted, uploads deleted after 24 hours, results after 72, no training on customer data.

Experience

In [4]:
zaki.experience()
Out[4]:
  1. Founder & Full-Stack Engineer

    /ForecastMyShop2026 – present

    Designed, built and launched a forecasting product for independent retailers. Details above.

  2. Data Scientist

    /SISQUAL WFM, PortoFeb 2025 – present
    • Build and maintain FastAPI endpoints for training, prediction, evaluation and model management in a production workforce-forecasting pipeline (SQL Server, AWS SageMaker, S3).
    • Benchmark PyTorch models and AutoGluon on accuracy and latency; cut WMAPE by 18%. Run A/B tests of candidate models against the production baseline.
    • Build feature pipelines with external signals (weather, football fixtures, holidays, events) and automate GPS event-data processing across client deployments.
    • Lead the DataHub rollout (domains, ownership, glossary, lineage, PII classification) and prototype MCP- and RAG-based assistants on a governed semantic layer.
    • Meet weekly with the Chief of Innovation and the Chief Developer to present results and shape the data and AI roadmap; run internal workshops and demos.
  3. Data Analyst

    /SISQUAL WFM, PortoAug 2023 – Feb 2025
    • Built Python and SQL ETL workflows and data-quality checks for forecasting and reporting.
    • Automated dashboards and reports, cutting manual querying and reporting time by about 40%.
  4. Computer Systems Engineer

    /Ministry of Higher Education, AfghanistanFeb 2019 – Aug 2021
    • Modernized HEMIS, the national higher-education management system (Laravel/PHP), used by 500+ university IT staff, and built its analytics and reporting modules.

Education and research

In [5]:
zaki.education(include_research=True)
Out[5]:
  1. MSc in Data Science and Engineering

    /University of Porto (FEUP)2022 – 2025

    Thesis: Improving Pharmacy Sales Prediction: A Global Forecasting Approach (dataset confidential)

  2. External Researcher

    /LIAAD, INESC TECAug 2023 – Oct 2024
    • Benchmarked DeepAR, N-BEATS and XGBoost global models against SARIMAX and Holt-Winters, using PySpark on 20M rows from 119 community pharmacies.
    • Improved accuracy by 9% over an autoregressive baseline through feature engineering and Optuna tuning.
    • Academic projects: CNN wind-turbine defect detection (ResNet, MobileNet, EfficientNet), LLaMA-based question answering, transformer sentiment analysis with Spark and CoreNLP.
  3. BSc in Computer Systems Engineering

    /UET Peshawar, Pakistan2014 – 2018

    Full scholarship awarded by the Prime Minister of Pakistan for international students.

// both degrees taught fully in English

Skills

In [6]:
print(zaki.skills)
forecasting
Chronos-2DeepARN-BEATSSARIMAXHolt-Wintersglobal modelsback-testingA/B testing
machine_learning
PyTorchXGBoostAutoGluonOptunascikit-learnCNNstransformers
data
PythonpandasNumPyPySparkSQLSQL ServerRPower BIPlotly
engineering
FastAPIREST APIsETL pipelinesLaravel/PHPHTML/CSSbilling & authi18n
mlops_cloud
AWS SageMakerS3Vertex AIDockerGitMLflowWeights & BiasesDatabricks ML
governance
DataHubmetadatalineagebusiness glossaryPII classification
generative_ai
RAGMCPLangChainLlamaIndexHugging FaceOpenAI APIClaude APIOllama

Volunteering

In [7]:
zaki.beyond_work()
Out[7]:

Volunteer

/European Solidarity Corps2022 – 2023

Mentored refugees and migrants and translated for them (Refugees Welcome, JRS, CPR). Co-organized a youth exchange in Bosnia and joined exchanges in France, Portugal and Spain.

Contact

In [8]:
zaki.contact() # happy to talk forecasting, ML or data