Microsoft Corporation
Nov 2017 – Jan 2023Data Specialist · Data Analyst II · Data Analyst
Microsoft India Development Center, Hyderabad, India
Data Sourcing Pipeline — Outlook Intelligent Cards
- ›Designed, built, and maintained an end-to-end data-sourcing pipeline for Outlook Intelligent Cards training data in a privacy-constrained ("eyes-off") environment, ingesting classifier-sampled emails into Cosmos and structuring them into queryable tables.
- ›Ran the pipeline as a daily batch job processing 1TB of data, engineering it to meet a 95%+ availability target for downstream model training.
Template Clustering & Extraction Modeling
- ›Designed a template-clustering system using HTML DOM pattern-matching to sub-classify sampled emails by sender/format template without access to raw content.
- ›Built high-accuracy data extraction and modeling pipelines achieving 95%+ precision and recall, scaling support to 100+ providers across travel, package delivery, and scheduling domains.
Live-Site Reliability & On-Call Operations
- ›Drove reliability as on-call engineer for live-site operations, using ICM for incident tracking and root-cause log analysis to maintain >99.9% availability.
- ›Led knowledge-sharing sessions and troubleshooting-guide updates to improve on-call documentation and speed up incident resolution.
PII Anonymization & Data Compliance
- ›Built a PII anonymization pipeline using HTML structural analysis to locate and replace sensitive fields with synthetic data, serving as the compliance gate transitioning data from eyes-off to eyes-on environments.
Cortana — Meeting-Scheduling NLU
- ›Improved Cortana's meeting-scheduling intent recognition to 90%+ accuracy by building and refining semantic parsing logic that extracts attendee, date/time, and location details.
Bing Search — Relevance Model Evaluation
- ›Collaborated with the Bing Search relevance team to evaluate ranking models across multiple search scenarios, using precision/recall analysis to identify failure patterns and improve model performance beyond the 90% production deployment threshold.
CosmosData PipelinesBatch ProcessingHTML DOM ParsingICMPII AnonymizationSemantic ParsingSearch RelevancePrecision/Recall Analysis