I Turn Data & AI
Into Business Results

Business Problem

Enterprise PostgreSQL databases contained critical operational data that AI agents couldn't reliably query — lacking schema documentation, metadata context, and a governed access layer. Analysts spent hours manually extracting data for ad-hoc queries.

Solution

Designed and built a RAG data-readiness framework that automatically documents schema structure, enriches table/column metadata, profiles data quality, and indexes this context into a vector database — enabling AI agents to query enterprise data through natural language with production-safe, governed access.

Architecture & Stack

• Automated schema documentation pipeline extracting table relationships, column types, and business context
• Metadata enrichment layer appending semantic descriptions to raw database objects
• Data profiling module assessing completeness, uniqueness, and freshness per table
• Vector embedding and indexing of schema+metadata for similarity search
• Governed query interface connecting AI agents to production data without direct DB access

Impact

What This Proves

Ability to design AI-ready data infrastructure: not just building AI tools, but preparing enterprise data to be safely and intelligently accessed by AI systems — a critical prerequisite for any serious AI deployment.

Business Problem

Logistics operations lacked a systematic way to diagnose shipment delays. Root-cause analysis was manual, slow, and inconsistent — preventing operations teams from acting on delay patterns before they escalated into SLA breaches.

Solution

Designed a Medallion Lakehouse on Microsoft Fabric ingesting logistics data across Bronze (raw), Silver (validated), and Gold (analytical) layers — with PySpark transformation logic and an LLM layer that analyses Gold-layer delay records to identify and explain root causes, surfaced in Power BI for operations leadership.

Architecture & Stack

• Bronze layer: raw logistics event ingestion via Fabric Dataflows
• Silver layer: PySpark validation, deduplication, schema enforcement
• Gold layer: aggregated shipment KPIs, delay flags, carrier performance
• LLM analysis layer: pattern-based delay root-cause generation from Gold data
• Power BI semantic model connected to Gold layer for real-time delay intelligence

What This Proves

End-to-end Microsoft Fabric Lakehouse implementation combined with applied LLM reasoning — demonstrating ability to build modern, AI-augmented supply chain intelligence platforms on the Microsoft data stack.

Business Problem

Freight and logistics operations had no unified view of carrier performance, route efficiency, or cost-per-shipment — forcing managers to manually compile data from multiple systems to answer basic operational questions.

Solution

Built a comprehensive freight intelligence dashboard consolidating shipment data, carrier SLAs, route performance, cold-chain compliance, and cost analytics into a single real-time Power BI platform — covering import/export operations, delivery against plan, and carrier benchmarking.

Key Features

• Real-time SLA tracking across carriers and route segments
• Cost-per-shipment and cost-per-km analysis by carrier and route
• On-time delivery performance vs. plan with variance analysis
• Cold-chain compliance monitoring with temperature-exceedance alerts
• Automated daily refresh via ETL pipeline eliminating manual data pulls
• Executive summary page with drill-through to route-level detail

Impact

What This Proves

Logistics domain expertise combined with Power BI/DAX proficiency — ability to translate complex freight operations data into clean, decision-ready analytics for operations and senior leadership.

Business Problem

One of Pakistan's largest logistics conglomerates had no centralised BI function — reporting was manual, took 2 days to produce, and lacked real-time IoT data for cold-chain compliance monitoring.

Solution

Built Gerry's entire BI and data function from scratch: designed and implemented ETL pipelines, a centralised data warehouse, Power BI dashboards covering logistics operations, and integrated real-time IoT sensor data for cold-chain spoilage monitoring. Also hired and mentored a 3–4 person analytics team.

Architecture & Stack

• ETL pipelines ingesting operational data from multiple source systems
• Centralised data warehouse with governed data model
• Real-time cold-chain IoT integration: temperature monitoring against SLA thresholds
• Power BI dashboards for logistics operations, route performance, and management reporting
• Automated daily reporting pipeline replacing 2-day manual cycle

Impact

Business Problem

A large national FMCG conglomerate with multi-facility operations and 12+ regional distributors had fragmented inventory, WIP, and supplier data across SAP — no consolidated view for planning or performance management, with manual reconciliation across dozens of data sources.

Solution

Built a consolidated national supply chain intelligence system extracting and transforming SAP S/4HANA operational data into a Power BI analytics layer covering inventory turns, WIP costs, material lead times, distributor reconciliation, and supplier KPIs — with automated reconciliation pipelines across 12+ distributors.

Key Capabilities Delivered

• National inventory, WIP, and lead-time dashboards with drill-down by facility, SKU, and distributor
• Automated reconciliation pipelines across 12+ regional distributors improving data quality
• Time-series demand forecasting models (SARIMA, ARIMA, Holt-Winters) for planning support
• Supplier performance KPIs feeding into S&OP process
• SAP S/4HANA MRP-aligned data extraction layer for analytics

Impact

Business Problem

Procurement teams needed a reliable forward cost view for petrol prices to inform purchase planning and budgeting decisions — currently relying on lagging spot-price data with no predictive outlook.

Solution

Benchmarked three statistical time-series forecasting models on 12 years of national petrol price data, selecting the best-performing model for production use — delivering a 4–6 week cost outlook that procurement teams could use directly for planning.

Methodology

• 12-year national petrol price dataset collection and preprocessing
• Stationarity testing (ADF), seasonal decomposition, ACF/PACF analysis
• SARIMA: seasonal parameter tuning (p,d,q)(P,D,Q)m grid search
• ARIMA: non-seasonal baseline with AIC/BIC optimisation
• Holt-Winters: triple exponential smoothing with additive/multiplicative seasonality
• Train/test split evaluation: MAPE, RMSE, MAE across all models
• SARIMA selected: 3.17% training MAPE, 15.65% testing MAPE

Impact

What This Proves

Deep time-series modelling capability — the ability to select, tune, and rigorously evaluate statistical forecasting models against a real business planning need, not just run a tutorial notebook.