As an Engineer on HubSpot’s core CRM relationships system, I work on some of the company’s most critical backend services. Our stack handles more than 26 billion requests per day while maintaining sub‑100 ms p99 latency and read-your-writes guarantees. This is backed by some of the largest HBase data clusters and Kafka clusters at the company. I focus on reliability, large scale distributed systems, and making it easier for product teams to build on top of shared CRM data.

Selected work:

* Led a zero‑downtime migration of core CRM relationship metadata from hard‑coded enums to a centralized, dynamic schema service, coordinating changes across 100+ codebases and 1000+ usages. This cut time‑to‑production for new relationship types from weeks to minutes and removed an entire class of rollback and scalability issues.

* Designed and shipped a new asynchronous write pipeline for high volume relationship updates on top of a wide‑column datastore. I worked around the lack of transactional secondary writes while preserving strict read‑your‑writes guarantees, and materially lowering pager noise for a tier‑1 service.

* Independently conceived, designed, and built an internal rate‑limiting debugging tool as a side project, using front‑end skills I developed on my own. The tool gives engineers real‑time visibility into throttling and has significantly reduced inbound questions and time to debug production issues.

* Act as a cross‑team technical lead for company‑wide bets, including an extensible app‑defined objects platform and a new work management experience used across multiple product areas. I own relationship modeling, permissions, and reliability contracts so new products can safely build on shared CRM data.

* Serve as a go‑to engineer for deep reliability and incident work, consistently root‑causing long‑standing race conditions and performance issues in distributed systems, then driving fixes from design through rollout while raising the bar on architectural rigor.

Backend Technical Lead for HubSpot’s Import and Export systems, leading a team of three engineers on high scale data ingestion and extraction. I owned technical direction, worked closely with product to shape the roadmap, and focused on reliability, scalability, and a great first-time data onboarding experience.

Selected Work:

* Proposed, prototyped, and led end-to-end delivery of pre-import data validation, initially built as a hackathon MVP to win leadership buy-in. The feature became one of the most used parts of the product, now validating roughly 20,000 imports per day, improving import accuracy by about 40%, earning an internal award, and helping customers start with clean CRM data.

* Architected and implemented a cross-product read-only pipeline that guarantees immutability of closed deals, including a major overhaul of the permissions system for granular object-level access control, while keeping the shared object-read service within strict latency SLOs.

* Designed and built invalid-enumeration remapping in the import flow so customers can map messy spreadsheet values to clean system values directly in the UI, without editing files by hand, significantly improving data quality at import time.

* Designed and implemented the Public Export API, focusing on a clean, evolvable surface for users, load testing to set safe rate limits, and swimlaning heavy workloads so performance issues are isolated from other teams and from UI-driven exports.

* Drove major reliability and performance improvements as volumes scaled, including rearchitecting a high-throughput cache to eliminate JVM out-of-memory issues and latency spikes, fixing a Kafka producer/consumer design that caused uneven partitioning and lock contention, and separating request schemas from storage schemas to avoid serialization breakage as the product evolved. I also documented our work on preventing serial processing in the import pipeline on the HubSpot engineering blog.

* Designed and built read-only pipeline stage based permissions. This was Hubspot's first granular permission. Previously permissions were closer to an all or nothing model, but with this feature users have different permissions based on an object being enrolled in some criteria. Implementing this feature involved me writing features for many different teams, including modifying our high volume object write services and designing this feature in a way that did not add latency in this write path that serves millions of requests per second at peak volume. I also had to ensure consistency in preventing these writes across the entire Hubspot platform.

Backend Engineer working on CRM imports, exports, and notifications.

Started on a team with me and just my tech lead (2 engineers). During that time I've helped with developing large customer facing features and improving the reliability and scalability of our systems. I've also helped mentor two engineers that joined the team since I started.

I've worked on a few large cross-company features:

* Backend implementation of Flexible Association support for CRM Imports: https://knowledge.hubspot.com/crm-setup/associations-enhancements-beta#import-to-set-company-associations-and-labels-in-bulk

* Adding Custom Object support for all CRM Notifications, including migrating data stores off of SQL via MapReduce jobs for improving API/Kafka performance:

https://knowledge.hubspot.com/crm-setup/use-custom-objects

* Migrating our internal Export Framework to use an improved object searching system

I've improved reliability of our systems while we scale:

* Solved major source of failed import requests resulting in elevated 500s due to downstream APIs

* Decreased API latency by improving overall performance of API via more asynchronous code

* Increased throughput of Kafka workers by adding caching layers, splitting workers up, and asynchronous code