Ruchir Kumbhare

3+ years of work experience

Oct 20241 y 11 moPresent

Bengaluru

◦ Connector SDK: Built a containerized Connector Framework for data ingestion adopted as the internal standard at Hevo — 10+ connectors in production, with all future connectors mandated to build on it, eliminating redundant infrastructure and standardizing ingestion across databases and SaaS sources. ◦ Built SQL Server Connector: Built a SQL Server connector on top of the Connector SDK using Change Tracking for incremental replication, achieving 1.52 GB/min throughput with low-latency syncs and minimal source overhead. ◦ Cost Optimization: Increased async process concurrency per instance, reducing autoscaled instance count and improving cluster CPU utilization from ∼15% to ∼50% (3.3x) across data ingestion workloads — cutting infrastructure spend by $105.6K annually with no impact on reliability or throughput. ◦ Improved Observability: Implemented ingestion throughput and offset movement metrics for SaaS connectors, enabling real-time pipeline monitoring and reducing mean time to detect and resolve ingestion failures. ◦ Critical Shopify Authentication Upgrade: Migrated 500+ customer pipelines from API key authentication to OAuth 2.0 within a 1-week deadline before Shopify’s deprecation cutoff — preventing live pipeline failures and unblocking new pipeline creation for all affected customers. ◦ AI Agent for Bug Detection: Built an agentic AI system using Claude Code to detect data-loss bugs in connector polling logic — grounds decisions on historical bug patterns, reads source code, and recreates crash scenarios; identified 10+ production bugs across 20 connectors and actively used by oncall engineers to accelerate RCA. ◦ VPC Peering Provisioning: Built cloud networking infrastructure using Pulumi to establish VPC peering between Hevo’s per-customer data plane and customer-owned cloud environments, enabling direct, secure database connectivity without public internet exposure for 100+ customer pipelines.

Jul 20243 moOct 2024

Bengaluru

DORA-Metrics: Built the end-to-end service for gathering, analyzing and presenting DORA Metrics. Set up integrations with Bitbucket, Jenkins, etc. to receive live engineering metrics. Utilized OpenSearch for storage and analytics.

Jan 20247 mosJul 2024

Bengaluru

◦ High Availability: Setup a docker compose for high availability of Tile38 geospatial database. Tested the various failover mechanisms for the same exhaustively. ◦ Testing Setup: Setup the testing mechanism(in-service tests) from scratch in a Kotlin Repo, with junit. Setup Integration testing for Jenkins Pipeline by picking up IP addresses of test Docker containers dynamically. ◦ REST APIs: Wrote multiple REST APIs pertaining to location services. Published and consumed HTTP Clients for the same. ◦ Firestore: Wrote connectors for firestore data uploading and consumption for location tracking service permissions. ◦ Location Tracking Improvement Taskforce: Part of the concerted effort to improve Location Recency of tracking experience by overhauling the infrastructure. Brought down the P95 Location latency from 2.44 mins to 0.19 mins.

Jul 20236 mosDec 2023

Hyderabad

Interned in the Passport Team within ShipTech org. ◦ Audit Trail: Developed the backend, using Java and SpringBoot, for writing REST APIs. Wrote Typescript code to manage AWS functionalities such as Lambda, DynamoDB, etc. ◦ Metrics: Worked on various packages to add incremental features such as metrics logging, configuring crawler jobs, adding lambdas, etc. ◦ Unit Tests: Wrote unit tests and ran integration tests on CI/CD pipelines ◦ Miscellaneous: Worked on bug fixes, service issues and version control.

May 20233 mosJul 2023

Remote

◦ LLM Research: Theorized and devised enhancements for existing open-source Large Language Models (LLMs) to optimize their performance and capabilities using PyTorch. ◦ Fine Tuning: Conducted fine-tuning procedures on LLMs, tailoring them to Machine Translation task to achieve higher accuracy and relevance. ◦ Benchmarking: Setup an evaluation pipeline, in Python, for LLMs to be evaluated objectively on their quality of Machine Translation.