Gaborone, Botswana • Data Engineer & AI Creator

Hi, I'm Sandile. I engineer data systems, build AI automations, and share the journey.

Business Intelligence Consultant & First-Class Honours graduate exploring the intersection of enterprise data architecture, practical Business AI Automation, and building a life in tech.

1.7B+ Records Automated across 3 countries
1st Class Honours BSc BI & Analytics (6 Distinctions)
30+ Pipelines & AI SQL, SSIS, n8n & AI Workflows
Kaggle & Python Predictive ML & modular pipelines
Now Building: Agentic ETL alert triage • Modular tabular pipelines on Kaggle • Filming Ep. 01 from Gaborone
Sandile Desmond Mfazi
Data • Predictive ML • AI Systems Enterprise BI at UGT, Predictive Modeling on Kaggle & launching my Business AI Automation YouTube channel
Production Data, Predictive ML & Business AI Stack Across Southern Africa
SQL ServerSQL Server SSISSSIS ETL SSASSSAS Tabular SSRSSSRS Reports Power BIPower BI & DAX Power AppsPower Apps Power AutomatePower Automate Azure Data FactoryAzure Data Factory Azure SQLAzure SQL Google BigQueryBigQuery LookerLooker Studio MariaDBMariaDB Oracle DBOracle DB SharePointSharePoint Microsoft ListsMicrosoft Lists ExcelExcel Modeling Kaggle • Python & scikit-learn n8n & LLM AI Agents
What I Do & Create

Bridging enterprise data engineering with predictive ML and AI automation.

By day I architect cross-border data warehouses across Southern Africa. Beyond that, I build modular Python pipelines on Kaggle, design Business AI Automations, and prepare tutorials for YouTube.

01 • Business AI Automation

AI Agents & Workflows

Designing practical AI agents, automated pipelines (Power Automate, Make, n8n, Python), and schema-validated LLM workflows that remove repetitive operational work.

n8n / Make Power Automate LLM Agents
02 • Data Engineering & Kaggle ML

Warehouses & Predictive Pipelines

Architecting 1.7B+ row SQL Server & Azure warehouses while sharpening exploratory data forensics and modular scikit-learn pipeline design on Kaggle.

SQL / SSIS / SSAS Kaggle & Python Power BI / DAX
03 • Personal Life & Creator Journey

Tech, Systems & Life

Documenting life in Gaborone, continuous craft after graduating First-Class Honours (6 distinctions), Notion systems, and building a creator studio from scratch.

Notion Certified Gaborone, BW YouTube Prep
System Architecture Blueprint

From disparate databases to executive BI and autonomous AI agents.

Click through the 4 engineering stages I use across banking, micro-lending, and modern AI workflows to turn fragmented data into trusted decisions.

Stage 01 • Operational Systems

Unified Extraction Across Legacy & Cloud Stores

Capturing high-frequency transactional records from core banking, micro-lending, ERP, and spreadsheet environments across Botswana, South Africa, and Zambia without impacting source performance.

SQL Server Azure SQL MariaDB Oracle DB Flat CSV / Excel
Input Sources Core Banking (BW/ZA/ZM) → Regional Loan Ledgers & CSVs
Capture Pattern Watermark Delta Extraction + Incremental Snapshot Checks
Reliability Target Zero Source Lockups • Deterministic Row-Count Parity
Stage 02 • Cleansing & Modular Logic

Deterministic Staging + Kaggle-Grade Python Pipelines

Standardizing schemas, resolving currency and date anomalies, and running modular transformation steps—combining enterprise SSIS & Azure Data Factory orchestration with Kaggle-honed Python preprocessing.

SSIS Packages Azure Data Factory Power Automate Python • scikit-learn
Staging Transformation Data Cleansing • Deduplication • Cross-Currency Normalization
Predictive & Forensic Layer Python ColumnTransformer • Gap Reconstruction Logic
Schedule Automated Nightly Batches • 30+ Reusable ETL Pipelines
Stage 03 • Semantic Foundation

Compressed Star Schemas & Tabular Analytical Cubes

Modeling fact and dimension tables at billion-row scale so leadership queries resolve in seconds rather than timing out, while providing structured single-source-of-truth context for AI agents.

28 SSAS Cubes Google BigQuery Governance Registers SharePoint Lists
Warehouse Scale 1.7B+ Records Stored Across 47 Compressed Tables
Dimensional Model Kimball Star Schema • SCD Type 2 Historical Tracking
Impact Reconstructed 115M+ Missing Historical Contract Records
Stage 04 • Decision & AI Action Layer

Executive Dashboards, Operational Apps & AI Workflows

Delivering KPI scorecards in Power BI and SSRS, corporate Risk Registers in Power Apps, and connecting clean warehouse triggers to n8n / LLM triage agents with strict human-in-the-loop audit logs.

Power BI & DAX SSRS Paginated Power Apps Looker Studio n8n / LLM Agents
Executive BI Margin, Default Risk & Multi-Country P&L Dashboards
Workflow & Risk Apps Automated Corporate Risk Registers & SLA Escalations
Agentic AI Bridge Structured JSON Schema → Bounded LLM Reasoning → SQL Audit
Kaggle Community • Predictive Analytics & Modular Design

My Kaggle Journey: Where Raw Data Meets Modular Python Pipelines

Beyond enterprise SQL warehouses, Kaggle is my open-source engineering gym—training fast exploratory forensics, leakage-free scikit-learn pipeline composition, and predictive modeling that feeds directly into real-world scorecards and AI systems.

Read Kaggle Essay Open Kaggle Lab →
01 • Forensic EDA Protocol

Treat Every CSV Like an Undocumented Production Feed

Before fitting a baseline model on Kaggle, I profile cardinality, null patterns, target imbalance, and temporal leaks—the exact discipline that later helped me reconstruct 115M+ missing historical contracts in SQL Server.

Missingness Heatmaps Leakage Audits pandas / numpy
profile = df.agg(["nunique", lambda s: s.isna().mean()]).T # Flag high-cardinality IDs & post-outcome columns before split
02 • Signal Creation

Domain Ratios Outperform Blind Hyperparameter Sweeps

Whether modeling credit repayment risk or food delivery order velocity, engineered ratios (utilization, rolling 30-day cadence, peer-group deviations) give interpretable lift across both linear models and gradient boosting.

Weight-of-Evidence Lag & Velocity Ratios Custom BaseEstimator
df["utilization_ratio"] = df["balance"] / (df["credit_limit"] + 1.0) df["tx_velocity_30d"] = df["tx_count_30d"] / (df["tenure_mo"] + 1.0)
03 • Modular Engineering

Composable Pipelines Without Notebook Spaghetti

Instead of mutating global DataFrames across 40 notebook cells, every imputer, scaler, encoder, and feature generator sits inside one reproducible Pipeline validated inside Stratified K-Fold CV.

ColumnTransformer StratifiedKFold CV Reproducible Artifacts
model_pipe = Pipeline([ ("features", DomainRatioFeatures()), ("prep", ColumnTransformer([("num", num_pipe, NUMS), ("cat", cat_pipe, CATS)])), ("clf", HistGradientBoostingClassifier(random_state=42)), ])
04 • Production Impact

From Python Notebooks to Live SQL Views & AI Triage

When stakeholders need transparent micro-lending scorecarding or AI agents need calibrated anomaly alerts, Kaggle-tested logic translates cleanly into SQL Server views and deterministic LLM routing thresholds.

Micro-Lending Scorecards T-SQL Translation AI Routing Gates
CREATE OR ALTER VIEW dbo.vw_CustomerRiskScorecard AS SELECT CustomerID, CASE WHEN UtilizationRatio > 0.82 AND MissedPay30d >= 1 THEN 'HIGH' ... END AS RiskTier FROM dbo.FactLoanBehavior;
The AI & Tech Blog

Latest Writing & Field Notes

View All Posts →
Channel In Production • Coming Soon

The YouTube Channel: Business AI Automation, Tech & Life

Inspired by practical, no-fluff AI builders like Nate Herk, I'm launching a YouTube channel dedicated to step-by-step Business AI Automation tutorials, real-world data engineering breakdowns, and honest stories about building a career in tech from Gaborone.

3 Core Channel Themes
  • 01. AI Automation Builds: Agents, n8n, Make, Power Automate & LLM workflows.
  • 02. Data & Tech Stack: SQL Server, Python, Power BI & cloud architecture.
  • 03. Personal & Systems: Notion setups, career lessons & life in Botswana.
Experience & Background

Career Snapshot

Full Interactive CV →
March 2025 – Present • Gaborone

Junior Business Intelligence Consultant

UGT — Analytics Training & Consulting

Engineered 30+ automated SSIS ETL pipelines and 28 SSAS Tabular Cubes, optimizing 1.7B+ records across 47 compressed tables for operations across Botswana, South Africa, and Zambia.

Apr 2024 – Nov 2024 • Remote

Power BI Expert & Data Analyst

Upwork — Verified Freelancer

Delivered custom BI solutions, DAX semantic modeling, data quality frameworks, 3-page executive KPI dashboards, and rapid SQL dataset analyses for international clients.

Feb 2023 – Aug 2023 • Gaborone

Business Intelligence Attachment

First National Bank of Botswana (FNBB)

Developed database tables and automated ETL processes using SQL Server Full Stack (SSIS, SSAS, SSMS) and Python, achieving 100% on-time delivery for paginated stakeholder reports.

Graduated 2024 • Gaborone

BSc (Hons) Business Intelligence & Data Analytics

Botswana Accountancy College / University of Sunderland

Awarded First-Class Honours (1:1) with all six distinctions, alongside Google Data Analytics and Notion Certified (Essentials, Workflows, Advanced) credentials.

Selected Work

Featured Projects & Deep Dives

Explore All Projects →
Enterprise Data Engineering

115M+ Record History Reconstruction

Parsed 115M+ flattened audit logs and automated 30+ SSIS pipelines across Botswana, South Africa, and Zambia with country-level Row-Level Security.

Read Case Study →
Serverless Cloud Portal

Real-Time Livestock Logistics App

Replaced vulnerable paper field logs with a serverless Google Apps Script, BigQuery, and Looker Studio pipeline syncing in under 60 seconds.

Read Case Study →
Interactive Lab

Governed Power BI Semantic Layer

Audited fragmented Excel metrics and unified Finance, Sales, and Marketing KPIs inside a documented, Git-ready Power BI Semantic Model.

Open Interactive Lab →
Free Weekly Notes & Templates

The AI & Systems Dispatch

Get practical Business AI Automation blueprints, SQL & Power BI engineering notes, and an alert the day my first YouTube videos go live.