Arman Vaziri

Arman Vaziri builds software and data systems that help growth and sales teams turn campaign ideas into coordinated execution. At the time of his World’s Fair 2026 appearance, he led Ramp’s product and sales-led growth engineering teams. His work there included automated outbound, an internal customer data platform, and Ramp Revenue, a team-built internal application.

From outbound automation to shared infrastructure

At Ramp, the automation effort began with smaller, specific problems. In his 2026 talk, he recalled working with one other engineer roughly three years earlier on automated outbound: gathering prospect data and using GPT-3.5 to generate personalized copy for sales sequences. Solving those immediate needs helped the team understand which data integrations and execution patterns could support other workflows. He advises smaller companies to follow that progression rather than spend a year designing a comprehensive architecture before delivering useful automation.

The problem driving this work is the coordination required after someone proposes a campaign. Teams must identify an audience, prepare materials, distribute instructions, and get busy sellers to execute them. Inconsistent customer records make coordinated targeting difficult, while meetings and administrative work leave little time for experimentation. Vaziri’s approach starts with shared data, develops useful applications for individual teams, and then connects the capabilities those applications establish.

The systems behind meeting preparation

Ramp’s internal customer data platform combines CRM records, product usage, enrichment, buying signals, and interactions such as emails, calls, meetings, and page views. A Postgres database maintains transactional guarantees and relationships between entities across systems, along with metadata about where information came from and when it arrived. Real-time events enter through infrastructure such as Kafka; enrichment jobs gather information about prospects and customers; and warehouse processing with dbt and Snowflake feeds computed data back into the operational platform through reverse ETL. These paths give agents access to customer context assembled from several sources.

• Meeting context: Pre-meeting briefs make the purpose of that infrastructure concrete. An account manager preparing for a customer meeting needs product usage and account health, the intended agenda, open support tickets, and recent customer requests. Before assembling that context, the system must identify which account the meeting concerns. Vaziri describes this as an unexpectedly difficult matching problem: the same attendee email can represent multiple businesses. The team maps meeting details to accounts and persists the association so downstream applications can reuse it.
• Reliable execution: The briefs run through durable agent execution built around Temporal. Each thread represents model and tool calls as activities, allowing execution to resume with its accumulated state after a worker failure instead of restarting the task. Agents receive tools scoped to their configuration, and workflows can pause for human input before continuing. For meeting preparation, background agents run nightly and combine the customer data platform, searchable unstructured information, system-level meeting-preparation skills, and employees’ custom instructions.
• Retrieval and customization: Vaziri places particular value on the information held in emails, call transcripts, notes, product knowledge, and sales playbooks. Ramp’s team chunks and embeds this material in Turbopuffer, then combines vector, keyword, and attribute search to retrieve what an agent needs—for example, information associated with a specific account. Selective retrieval avoids placing the entire corpus in the model’s context. A customizable skill library lets employees describe which information and format they want, making briefs fit their working habits. Vaziri credits that customization with helping adoption.

Extending workflows and learning from employees

Those capabilities can support additional work, but extending them requires understanding the next user’s needs. Account executives dealing with prospects need more third-party information than account managers supporting existing customers, so the data integrations and skills must change. Vaziri also describes potential extensions to post-meeting follow-ups and CRM updates: an agent could identify an expansion opportunity in a transcript, prefill the required information, and ask the seller to approve creating it.

Ramp exposes the same tools and skills used by background agents through an internal Model Context Protocol server. Employees can query customer information, build agents, and create their own automations. Their prompts, skills, and working applications reveal specific problems the engineering team can help turn into supported tools for others. This gives Vaziri’s team a way to discover useful workflows through what employees actually build.

Toward go-to-market orchestration

His larger ambition is go-to-market orchestration: describing a campaign’s intent and generating coordinated outputs across sales outreach, advertising, and the web. His example offers Pro V1 golf balls to golfers at East Coast construction companies. Executing that idea requires a consistent audience, personalized email sequences, landing pages, and advertising creative. The proposed orchestration would route those outputs through Ramp Revenue and the specialized systems already built, with channel owners reviewing and signing off. Vaziri presents that broader coordination as a goal built on existing narrower automations, with guardrails for compliance, rules of engagement, and avoiding repeated actions.