Senior Software Engineer @ Procol
Dec 2024 – Present
Leading the architecture of Clara, an AI agent that automates procurement workflows across Procol and third-party platforms like Ariba, Coupa, and SAP.
The problem
Enterprise procurement workflows are fragmented across Procol, Ariba, Coupa, and SAP. Every new integration required bespoke point-to-point plumbing, which slowed delivery and made the system brittle as the adapter count grew.
The approach
- Introduced an event-driven mediator pattern so producers stay decoupled from ever-changing platform adapters.
- Used MCP (Model Context Protocol) as a standard integration surface for tools and platforms.
- Built a dedicated AI orchestration service that decides what to publish, where, and when.
- Designed a Long-Term Memory Store so Clara carries episodic context across sessions.
- Tuned Puma, workers, and memory configuration to push throughput from 200 to 500 RPM while cutting server costs by 25%.
Impact
Reduced integration turnaround time for enterprise clients
Enabled multi-platform publishing from a single source of truth
Agent sessions now remember context across conversations
25% server cost reduction with 2.5× throughput increase
What I learned
- A mediator earns its complexity once you have more than two adapters with unstable contracts.
- Episodic memory changes user expectations — users start treating the agent like a colleague, not a search box.
- Throughput bottlenecks are often in the runtime configuration, not the language or framework.
Tech
Ruby on RailsMCPAI ServicesPostgreSQLRedisEvent-Driven Architecture