Ramp Feature Trails

Feature trailActive

Background Agent System

This post details the development and implementation of OCA, an AI on-call assistant built on Ramp's custom background agent system. It focuses on how OCA leverages AI to automate on-call tasks, reduce human intervention, and improve incident response times. This builds upon the foundational work of the background agent system by showcasing a specific, AI-driven application for operational efficiency. The system is designed to handle asynchronous tasks, manage state, and integrate with various internal tools and external services to automate complex workflows.

2 posts · 2026

Timeline

Feature trailActive

Cost Efficient Snowflake CI

This post introduces the use of Apache Arrow to significantly reduce memory consumption during data fetches from Snowflake, achieving up to 79% reduction. This builds upon existing efforts to optimize CI costs with Snowflake by directly addressing memory usage during data retrieval, a key factor in compute efficiency and cost.

2 posts · 2022–2026

Timeline

20222026

Feature trailActive

LLM Routing Optimization

This post introduces online learning techniques to optimize LLM routing for cost efficiency, enabling dynamic model selection based on real-time cost and performance metrics. It details how Ramp accelerated machine learning development to simplify finance by building a system that dynamically routes requests to the most cost-effective and performant ML models. This builds upon the previous iteration by incorporating online learning for dynamic, real-time optimization.

2 posts · 2023–2026

Timeline

20232026

Feature trailActive

AI Token Spend Pipeline

This post details the development of a unified pipeline for tracking and managing AI token spend. It details the technical architecture and implementation designed to provide a centralized view of AI resource consumption, enabling better cost control and optimization. This post further elaborates on the challenges of AI spend, highlighting the dual problem of overspending on underutilized models and underspending due to lack of awareness or access to more performant models. It proposes strategies to address these challenges by providing granular visibility into token usage across different models and use cases, enabling more informed decision-making for resource allocation and cost optimization.

2 posts · 2026

Timeline

Feature trailActive

Stack Benchmarking

This post introduces the development of a comprehensive stack benchmarking system at Ramp. It details the technical architecture, methodologies, and tools employed to measure and compare the performance of different components and configurations within Ramp's technology stack. This enables data-driven decisions for performance optimization and infrastructure upgrades.

1 post · 2026

Timeline

Feature trailActive

Automated Receipt Collection

This post details the automation of receipt collection using Apple Intelligence for on-device inference. It details the technical approach to leverage on-device ML capabilities for processing and categorizing receipts, aiming to improve efficiency and user experience in financial management. This builds upon the foundational work of the automated receipt collection system by integrating on-device inference capabilities.

1 post · 2026

Timeline

Feature trailActive

Automated Security Issue Remediation

This post details the development and implementation of an automated system for identifying and fixing security vulnerabilities, focusing on just-in-time access controls to balance speed and security. This enhances the existing automated security issue remediation capabilities by introducing fine-grained, time-bound access to cloud resources, and now includes proactive, automated remediation of ~100 security issues within a 6-day period with zero human intervention.

3 posts · 2023–2026

Timeline

20232026

Feature trailCooling

Industry Classification System

This post details the revamp of Ramp's industry classification system, moving from a RAG (Retrieval Augmented Generation) approach to a more robust and enriched system. This involved developing new models and data pipelines to improve the accuracy and depth of industry categorization for financial data. Specifically, it introduces the use of transaction embeddings to enhance retrieval accuracy, building upon the previous iteration of the industry classification system by leveraging AI to improve retrieval.

3 posts · 2024–2025

Timeline

20242026

Feature trailCooling

Ramp MCP Server

This post details the journey of building and scaling Ramp Bill Pay from its inception to handling significant volume. It covers the architectural decisions, technical challenges, and iterative improvements made to ensure reliability, scalability, and efficiency in processing bill payments. This builds upon the foundational work of the Ramp MCP Server by detailing the specific implementation and evolution of a critical product capability within that system.

3 posts · 2023–2025

Timeline

20232026

Feature trailQuiet since 2024

Python Codebase Modernization

This post details the process of modernizing Ramp's Python codebase to increase development velocity. It covers the technical strategies and architectural changes implemented to achieve this goal, focusing on improving code quality, maintainability, and developer productivity.

1 post · 2024

Timeline

20242026

Feature trailQuiet since 2023

Ramp Growth Engine

This post details the learnings from building and scaling Ramp's growth engine, focusing on the technical architecture, systems, and strategies employed to drive user acquisition and engagement. It builds upon the foundational work of the Ramp Growth Engine by sharing key insights and lessons learned during its development and scaling phases.

1 post · 2023

Timeline

20232026

Feature trailQuiet since 2022

Rate Limiting System

This post details the implementation of a rate limiting system using Redis, focusing on the technical challenges and solutions for managing API traffic and preventing abuse. It covers the architecture, algorithms, and trade-offs involved in building a robust rate limiting solution.

1 post · 2022

Timeline

20222026

Feature trailQuiet since 2022

Pull Request Merge Performance

This post details improvements to the pull request merge process, aiming to reduce merge times and improve developer productivity. This builds upon the foundational work of the Pull Request Merge Performance initiative by detailing specific optimizations and their impact.

2 posts · 2022

Timeline

20222026

Feature trailQuiet since 2021

UI Component Library

This post introduces the bootstrapping of Ramp's UI component library, focusing on the initial technical decisions, architecture, and development process to establish a foundational set of reusable UI elements for internal product development. This builds upon the foundational work of the UI Component Library by detailing the initial bootstrapping process.

1 post · 2021

Timeline

20212026

Feature trailQuiet since 2021

Elixir Adoption and Systems

This post introduces the adoption of Elixir at Ramp, detailing the initial motivations, technical considerations, and early experiences with integrating Elixir into the engineering stack. It lays the groundwork for future posts on how Elixir is used to build and scale specific systems.

2 posts · 2021

Timeline

20212026