Muhammad Usman Jehangir

Summary

What I did

• Designed and owned a multi-stage prospect enrichment pipeline on AWS Step Functions, Lambda, SQS and ECS, with PostgreSQL sub-step/action level execution logs and failed steps' S3 artifacts.
• Cut the pipeline's error rate from roughly 60% to 1-2% through structured experiments and scaled coverage from 30% to 90% in 30 days using capacity modelling.
• Designed an S3-backed cache with diff-based write-back for a rate-limited data source, cutting both development cycles and production execution times. The pattern was adopted across the repository.
• Built a React monitoring interface for pipeline health, data freshness and error rates.
• Built dependency-aware circuit breakers that re-evaluate pipeline health periodically. Opening an upstream circuit also opened its dependents to prevent needless wasting of resources.
• Built a config-driven KPI metrics engine, reducing the time to add and configure a new KPI from about an hour to under five minutes. Analyses based on these metrics helped with compensation policy decisions for a roughly 50-person sales floor.
• Built a lead-assignment engine informed by lead lifecycle analysis; qualification rates rose from roughly 3% to 7-8% across the sales floor.
• Built an AI-assisted class-code review workflow: historical approvals and declines guided recommendations with reasoning for new batches; a human reviewed them before persistence.
• Developed prospecting and lead generation pipelines to source and enrich business data for sales qualification.
• Built internal data pipelines for dashboards tracking sales KPIs, targets, commissions, and lead assignment.
• Developed an AI-assisted class-code review workflow using a phraseology-first approach to determine approval status.
• Designed the architecture for the classification system of the class code workflow.
• Built end-to-end reliable and scalable data pipelines in production.

Results

• Built a standardized loader for fetching data from Airtable that cached data locally in pkl files to reduce development and testing time.
• Expanded the Airtable caching system to production using S3 JSON storage to minimize API requests by performing operations locally before syncing.
• Cut pipeline error rate from 60% to 1-2%.
• Reduced KPI configuration time from 1 hour to under 5 minutes.
• Increased qualification rates from 3% to 7-8%.