Experience
Promart
Visual and semantic search for 1M+ products, estimated 83% potential infrastructure cost saving across 37 stores, plus Airflow-based analytics and pipeline migration.
The Promart tenure combines an early analytics automation phase with a later data engineering role focused on visual-semantic search and platform migration. Together they show the progression from instrumentation and reporting to search infrastructure, orchestration, and cost-efficient production systems.
Built semantic visual search infrastructure for 1M+ products across 37 stores using ResNet pooling-layer embeddings, pgvector/HNSW, and vector similarity search.
Replaced AWS CloudSearch with Typesense on Kubernetes, redesigning product/store data and estimating 83% potential infrastructure cost saving with similar latency and performance.
Migrated 50+ production DAGs from StreamSets to Airflow, cutting pipeline maintenance about 40% and improving extraction up to 10x with ConnectorX versus pandas read_sql.
Stabilised the OneLogistics restocking pipeline with API/database extraction and exponential backoff, reducing failures from about 1 in 5 runs to 1 in 134.
Earlier in the tenure, automated daily data collection and user event tracking with Airflow, Google Analytics, Looker Studio, and Python to improve product and marketing analytics.
1M+
Products supported
83%
Potential cost saving
37
Stores deployed
Airflow
Platform migration