Building B2B SaaS again (after Stripe), but now with AI agents. Really interesting technical challenges in both product and infra!

Worked as a tech lead on Stripe's Revenue Platform / User Billing system — a mission-critical & data-intensive distributed system in an extremely complex business domain!

We own one of the largest (in terms of throughput) and one of the most critical online data platforms at Stripe, the real-time revenue source of truth and attribution engine for the entire Stripe suite of products, down to the transaction-level accuracy and explainability.

Some highlights:

• Led a P0 effort to re-architect the low-latency revenue attribution engine's architecture (15k+ RPS) and data interfaces to mitigate a existential scalability bottleneck — collaborated with internal data infra teams, led technical discussions with multiple staff+ engineers across the data infra org.

• Led an effort to re-architect Stripe's invoicing pipeline (a P0 system that debits billions of fee records a day). Increased the system throughput by 100x — eliminated the scalability bottleneck which de-risked onboarding many high-profile large enterprise Stripe customers.

• Owned a critical slice of Stripe's accounting SOX compliance roadmap. Worked directly with internal and external stakeholders (Deloitte, EY) to scope out and implement SOX controls.

Stripe Connect / Accounts API and Infrastructure

Some highlights:

• Worked closely with end users to roll out multiple breaking changes in order to make Stripe Connect compliant with India Data Locality laws — unblocking Stripe's launch in India

• Productionized accounts-srv, one of Stripe's first high-throughput, low-latency JVM-based micro-services.

• Managed and mentored an intern — he successfully completed the intern project, earned an return offer, and eventually accepted the offer.

Worked on Stripe's KYC / Compliance Platform — the platform / rule engine that allows legals to express specific compliance rules for a particular financial product (eg. Stripe Issuing) in a simple DSL, which gets translated into Stripe merchant onboarding requirements and risk rules.

The domain was extremely complex — sitting in the middle of legal, product launches, risk enforcement... Had the pleasure to work with Drew and learned a lot from him (https://www.linkedin.com/in/drewhoskins2/).

I worked on the DSL authoring interface — came up with and rolled out a new abstraction that significantly simplified the compliance plan authoring experience.

My internship designation was GE (greatly exceeded expectations), highest performance rating for interns (top 2%)

Worked on Ads Events Infrastructure. Contributed to a revenue-critical stream processing system as part of Facebook's Ads feedback loop (impression/conversion -> training -> serving).

Specifically, I migrated a revenue-critical and high-throughput (1.5GB/s) C++ streaming data flow from an untyped JSON-encoded stream, to a typed schema-aware Thrift-based data stream.

This was one of those change-the-engine-while-the-plane-is-flying jobs. The difficulty was way beyond an intern project, but I still managed to complete the migration without incidents (surprised and delighted the team). This decreased CPU utilization by 22% and eliminated significant tech debt (stringly to strongly typed).

Specifically:

(1) Independently proposed and executed a significant refactor of the downstream stream processor to move its deserialization logic behind an interface (touched 8k+ lines of C++). This made the migration a lot more testable and much safer.

(2) Scoped, designed, and implemented a custom testing system to detect discrepancies between the two data flows before shifting traffic to the new one.

(3) Rolled out (in phases) the new data flow two weeks before my internship ended without causing any outages or revenue loss.