Alexander Shevchenko
Head of Applied Research, Ramp Labs at Ramp
About Alexander Shevchenko
Alexander Shevchenko leads Ramp Labs, the applied research group at Ramp, the corporate card and finance automation company. He joined Ramp's applied AI team about three years ago and worked across several products that apply LLMs and machine learning. Ramp Labs is behind Ramp Sheets, a spreadsheet agent for finance teams launched in November 2025.
In an AI Engineer talk, Shevchenko explains why an application company should do its own research even as frontier models get cheaper, faster and better.
Career Highlights
- Head of applied research at Ramp, leading Ramp Labs
- Member of Ramp's applied AI team since about 2023
- Behind Ramp Sheets, an early spreadsheet agent for finance work
- Speaker at AI Engineer (2026) on specialist models, cartridges and agent swarms
Notable Positions
On Specialist Models Under Frontier Orchestrators
Ramp kept Opus 4.6 as the planner in Ramp Sheets and handed financial retrieval and question answering to a Qwen 3.5 35B model post-trained with Prime Intellect. It beat Opus 4.6 on exact task accuracy at much lower latency. See Specialist Agents.
On Precomputing Context
Finance work such as month-end close is predictable, so Ramp Labs explores cartridges that compress 20 to 30 times more context into the KV cache overnight, before the user opens the agent.
On Agent Swarms
Ramp Labs used 10,000 parallel agents to pen test Ramp and found seven highest-severity vulnerabilities. Open-weight models up to five times cheaper came close to frontier detection rates in the swarm. See Model Routing.
Key Quotes
- "Over the last year, the number of use cases that use LLMs has increased 123%." (on AI use at Ramp)
- "No single component of that stack was unsafe." (on a vulnerability only an end-to-end swarm found)
Related Reading
- Fine-Tuning - Post-training a smaller model for a narrow task
- Specialist Agents - Narrow agents working under an orchestrator