Six years in data. CDC runtime down 67%. repo-context-hooks ships on PyPI.

Start with repo-context-hooks to see how I ship Python: 330+ tests, Sigstore-signed, at pypi.org/project/repo-context-hooks. fraud-detection runs on synthetic data, so judge the pipeline design and not the 2% fraud rate.

Three years in commercial ops taught me what bad data costs. I design for recovery: MTTR under an hour.

I run CI/CD through Azure DevOps: the cycle went from 3 months to 14 days. That covers 2 data pipeline jobs and 5 deployment pipelines, including the database copydown tool.

• I own Payroll CDC end to end: a failed load reruns itself, no manual cleanup.

• repo-context-hooks: I keep this on PyPI. Agent memory in a checked-in file, zero dependencies, 330+ tests, Sigstore-signed.
• autoapply-ai: 11 ATS adapters, 40+ endpoints, 355 tests, private.
• tailor-resume: bullets rewritten against a PDF-calibrated budget. 190 tests, PyPI.
• job-scout: 109 pages, per-scraper circuit breakers. One break costs 1 source.
• fraud-detection-ml-platform: Kafka scoring over 100K synthetic transactions. P99 1.12ms, 100+ TPS.
• FinTune: Mistral-7B QLoRA, PII redaction, drift monitoring. 35+ tests.

At Zomato I worked unit economics across 300 restaurants, where per-order contribution moved from -Rs18 to +Rs2. At udaan I built the BI tracking Rs5 Cr monthly GMV in three months.

Design docs for two attestation systems, PaySchedule and PayHoliday, plus business continuity runbooks and Azure visibility dashboards. New hires read my docs on code spaces, git branch strategy, and work item resolution instead of asking me.

• M.S. Information Science and Technology, Missouri S&T, 4.0 GPA
• DP-700, Microsoft Certified Fabric Data Engineer Associate
• Sentiment analysis paper, Journal of Nonprofit and Public Sector Marketing, 2025: doi.org/10.1080/10495142.2025.2525123
• narendranathe.github.io | substack | linkedin | edara.narendranath@gmail.com