Skip to main contentSkip to navigation

Syed Ahmed Basharat Ali

Projects

Smart Retail Operations Platform

By Syed Ahmed Basharat Ali.

An edge computer-vision and data platform that turns in-store CCTV into retail operations data: visitor counting, task management and real-time violation detection on devices inside each branch.

Challenge

The platform runs computer-vision inference on edge devices and moves the results through several systems before anyone sees a dashboard.

The hardest problems were the ones that did not fail loudly: pipelines that reported success while upstream data was missing, and a warehouse sync that kept timing out.

Approach

  • Traced an 80% upstream data loss across a 150-device fleet. Built reconciliation and monitoring first, so every hop could be measured, then shipped the fix as a staged, reversible rollout verified against the reconciliation numbers.
  • Stabilized a 137 GB PostgreSQL backbone that was failing 8-hour Airbyte syncs.
  • Shipped an 11-model dbt migration across a multi-tenant Snowflake warehouse (50+ client schemas), with reconciliation and observability built from zero in Python and GitHub Actions. It surfaced two real sync issues on its first run.
  • Violation rules resolve through a Branch, Brand, Abstract hierarchy where the most specific rule wins, so a location or brand can override platform defaults without touching shared logic.

Pipeline

  1. Edge: Jetson Nano, YOLO, OpenCV. On-device CV inference.
  2. Ingestion: Supabase (PostgreSQL, Edge Functions). Ingestion and real-time rules.
  3. Sync: Airbyte. Incremental sync to the warehouse.
  4. Warehouse: Snowflake. Multi-tenant storage and compute.
  5. Modeling: dbt. Tested models.
  6. BI: Metabase, Power BI. Dashboards and reporting.

Results

  • An 80% upstream data loss traced to its source and fixed in stages.
  • Warehouse syncs stabilized after the 137 GB PostgreSQL backbone was fixed.
  • An 11-model dbt migration shipped across 50+ client schemas, with reconciliation that caught two real sync issues on its first run.

Technologies

Jetson Nano, YOLO, OpenCV, SQLite, Python, Supabase, PostgreSQL, Airbyte, Snowflake, dbt, Metabase, Power BI, GitHub Actions.

Impact

The recovered data mattered, and the observability layer mattered more: it is the difference between a fix you hope worked and one you can show worked.