How Geneva Trading Slashed Token Usage and Gained 100% Hit-Rate on Trader Queries Using Driver

About

Geneva Trading is a leading proprietary trading firm with a storied history of consistent success in the listed derivatives markets. Over the past 25+ years, Geneva Trading has grown significant capital and developed proven technology while maintaining an appetite for diversified trading strategies.

---

Challenge

Bridging the Context Chasm Between Agentic Tools and Geneva’s Legacy Codebase

Geneva serves high-profile traders competing in global financial markets where execution speed and precision are key. Every capability built into its trading platform has market-facing impact: features that ship sooner open new opportunities, and features that stall do not.

For Sherwin Hamidi, Technology Lead for Geneva’s biggest market-making desk, the pressure to ship faster with a lean, seven-person engineering team was constant. So when agentic tools began proving themselves in greenfield projects, he theorized they could close the gap between what Geneva’s traders needed and what the team had the capacity to build. The problem? A looming context chasm between Geneva’s agentic tools and its nearly decade-old codebase.

“Every time we wanted to start a project, we would have to write this massive document explaining our codebase to the agent,” Sherwin explains. “It’d be this multi-step process before development could even begin.” As new projects quickly turned into briefing exercises, features that included cross-team orientation, unfamiliar codebases, or extensive setup overhead were quietly deprioritized because the upfront cost made them not worth starting.

This context chasm wasn’t isolated to internal workflows. Geneva’s trader-facing chatbot, built to field questions about the platform’s features and functionality, also routinely missed the mark, returning slow, noisy, or misleading answers rather than the crisp guidance users needed. “At times, it would simply make up answers,” Sherwin says. “I wouldn’t even necessarily call them hallucinations; it was just misinformation. And in trading, fabricated answers break trust.” As confidence in the chatbot dwindled, Sherwin found himself pulling senior engineers off active development to assist traders with queries.

Knowing these manual workflows weren’t scalable, Sherwin sought a context infrastructure platform that would unlock the true potential of Geneva’s agentic tools. When Geneva’s CEO suggested Sherwin meet with Driver, their underlying vision matched his problem exactly: keep platform knowledge current, make it available to agents and traders, and stop rebuilding context from scratch. He was in.

Context used to be a moving target that required constant manual upkeep. Driver completely automates that lifecycle, delivering highly accurate, rock-solid context in the background with zero management overhead from our team.

---

Solution

Pre-Compiled Context Infrastructure Gives Agents Complete Platform Knowledge Before the First Line of Code

With Driver, Geneva solidified a pre-compiled context layer for its agentic SDLC and trader-facing chatbot. Agents no longer start cold, developers no longer rebuild task briefings by hand, and traders receive answers grounded in accurate, always-current platform knowledge.

During onboarding, Driver worked through single-tenant architecture, private model endpoint planning, on-prem Bitbucket connectivity, Entra-driven permission inheritance, audit logging, and prompt-level audit requirements. As Sherwin put it, “They took on a huge lift to accommodate our security requirements.”

Today, Geneva’s front-office strategy team uses Driver’s context layer and Claude Code on real Jira tickets across multiple users and repos. Driver compiles architecture, dependencies, call paths, and file-level documentation ahead of time, then serves it through MCP. Now, Sherwin’s team no longer pays the context tax at the start of every project. No briefing documents, no manual file gathering, no re-orienting the agent from scratch, no hours lost exploring files. The work that used to begin with setup now begins with development.

That same architecture pays dividends on trader-facing queries. With Driver, they can query the platform directly: how does this feature work, what does it do, how do the configurations interact? “You jump on the chatbot and ask, and it does a beautiful job of explaining the whole thing,” Sherwin says. No senior engineer interrupted. No capability left unused.

The clearest breakthrough, however, came when Sherwin’s team needed a feature that touched Geneva’s market data codebase. Historically, that meant lobbying the market data team for roadmap priority and waiting for their capacity. With Driver powering Claude Code, Sherwin’s team pointed the agent at Driver-compiled market data context, described the change, and moved across the market data code, their own code, the GUI, and related components. The work that would’ve taken three to four months before was done in less than a week.

Driver’s token efficiency is a massive step up from our old Explorer tool. The drop in token burn is immediately obvious when we compare the usage data.

---

Results

Geneva Redefines What a Seven-Person Engineering Team Can Achieve With Driver

With Driver, Geneva moved agentic coding from isolated greenfield wins into production-grade workflows without friction. Sherwin’s team can now take on cross-team projects that previously required roadmap negotiation, while traders receive up-to-date platform knowledge without leaning on senior engineering resources.

• 100% hit-rate on all trader queries
• 50% reduction in token usage

Next, Sherwin’s focus is on operational leverage. He’s especially eager to expand coverage into broader markets and increase the platform’s reliance on automation, enabling the same desk to run more strategies without adding headcount.

The operational leverage we’ve gained with Driver is a game-changer. We’re aggressively tackling high-impact features that we previously had to deprioritize due to legacy overhead, and we’re executing them at a fraction of the time.

---

See how Driver helps AI coding agents work confidently across legacy codebases — with architecture, dependencies, and history compiled ahead of time.