Dhruv Mehra

Led end-to-end design and delivery of an EV rental microservice (Flask + DynamoDB + Kafka + AWS Lambda/ECS) serving 2.5L+ riders across 3,000+ dispatch centers — sole backend owner from scoping to production. Built an event-driven vendor inventory sync pipeline (S3 pre-signed upload → Lambda validation with row-level error reporting → batch DynamoDB upserts with exponential backoff → SES notifications). Designed and built an automated service dependency knowledge graph over 200+ microservices using Neo4j, ingesting from CD webhooks, static code analysis, and runtime HTTP call graphs via CubeAPM PromQL — enabling sub-second topology queries that previously required manual investigation across multiple dashboards.

Software Development Engineer 1 at Delhivery (2025-11 – Present)

• Led end-to-end design and delivery of an EV rental microservice (Flask + DynamoDB + Kafka + AWS Lambda/ECS) serving 2.5L+ riders across 3,000+ dispatch centers — sole backend owner from scoping to production.
• Built an event-driven vendor inventory sync pipeline (S3 pre-signed upload → Lambda validation with row-level error reporting → batch DynamoDB upserts with exponential backoff → SES notifications).
• Designed and built an automated service dependency knowledge graph over 200+ microservices using Neo4j, ingesting from CD webhooks, static code analysis, and runtime HTTP call graphs via CubeAPM PromQL — enabling sub-second topology queries that previously required manual investigation across multiple dashboards.
• Architected a multi-agent Root Cause Analysis system wiring LLM reasoning to Neo4j topology, CubeAPM PromQL metrics, and logs — reducing MTTR by ∼ 75% (avg. 45 min → ∼ 10 min) with ∼ 80% accuracy in automated failure classification across production incidents.
• Optimized Elasticsearch performance — implemented a singleton-pattern connection manager cutting redundant overhead, and migrated from must match to term filter queries reducing latency by ∼ 70% (45s → 14s) on high-volume listing APIs.
• Migrated multiple services to AWS Graviton (ARM64) via multi-architecture Docker images (linux/amd64 + linux/arm64), reducing EC2 compute costs by ∼20%.

Software Development Engineer 1 at Park+ (2025-08 – 2025-11)

• Architected and led the development of a real-time user segmentation engine that processes 40M+ frontend events and 30K+ transactional events daily maintaining accurate behavioral lifecycle stages (C0,L1,L2 etc) for ∼6 lakh monthly active users across 7 product verticals.
• Built a configurable recursive rule engine, enabling product and growth teams to define and modify user journey progression via JSON configs without code changes or deployments.
• Engineered an Elasticsearch sync service consuming 50K+ Kafka events per batch and indexing documents with dynamic time-based indices, enabling sub-second search latency for time-series analytics.
• Reduced database load by replacing unconditional stage lookups with targeted queries — cut daily MySQL reads from ∼3.8M to ∼400K, dropping peak QPS from ∼50K to ∼5K.
• Led the migration of Debezium Kafka Connectors to Kubernetes, reducing maintenance downtime.

Software Engineer Intern at Park+ (2025-05 – 2025-08)

• Developed a document storage API (Django REST Framework) for user KYC and profile images with GCS signed URLs and CDN access.
• Developed REST APIs for Kafka Connectors, enabling self-serve management of Debezium connectors without manual Kafka cluster access.
• Designed and deployed a Kafka-to-S3 ingestion pipeline processing 1M+ log events daily with Parquet compression and partitioning, improving Athena query performance by 40%.

B.Tech in Electrical Engineering – Delhi Technological University (DTU) (2021 – 2025)