Ramp applied artificial intelligence engineer Alexander Shevchenko demonstrated how targeted research outperforms frontier models in finance. Ramp trained a 35 billion parameter model that beat Opus 4.6 on financial spreadsheet tasks with lower costs and faster response times.