Ryan Clay

• •Built customer-facing AI agent capabilities across Zendesk's Resolution Platform — AI Agents, Agent Builder, and custom agents that automate support over messaging, email, and voice — working across the stack from React admin surfaces to Python agent-runtime services, and helping grow automated resolutions past 40% of eligible conversations for enterprise accounts.

• •Designed the multi-turn evaluation and semantic-observability layer that gates every agent release: accuracy/coherence/tone/safety scorers, adversarial multi-turn simulation, and automated regression detection — cutting production regressions ~35% release-over-release.

• • Built the RAG pipeline behind grounded answers — knowledge connectors (Help Center, web crawler, Confluence, Google Drive), retrieval with re-ranking, and locale/search-rule scoping over an in-house retrieval index — improving answer relevance ~22% vs. a keyword-search baseline (measured on an ~800-query labeled eval set) while reducing median response latency ~18%.

• • Implemented multi-model LLM routing and provider failover across OpenAI, Anthropic, Google, and Amazon Bedrock, selecting models per task on quality, latency, and cost, with multi-homing that kept agent availability above 99.9% during provider incidents.

• • Cut model cost per resolved conversation ~45% through task-based model routing, response caching, and prompt/context trimming — without regressing resolution quality.

• •Built a Python SDK and plugin interfaces plus a webhook-first, event-driven runtime (Chat, Ticket, and Widget Escalation APIs) for asynchronous agent orchestration and human handoff, using Pydantic typed schemas to normalize and validate agent execution and evaluation data.

• •Delivered action-flow and MCP tool integrations (Salesforce, Slack, Jira, Google Sheets) so agents could take authorized actions, enforcing deny-by-default tool access with read/write scoping to meet enterprise governance requirements.

• •Led backend and full-stack development for enterprise platform and workflow-automation features — Interface Designer, the Automations engine, the Connected Apps Platform, and admin/security tooling — used by thousands of organizations.

• •Launched enterprise governance capabilities including the redesigned Admin Panel, user groups, bulk permission management, audit logs, and workspace sharing restrictions, plus the early AI field type introduced in 2023.

• •Built the automation execution system — triggers/actions, webhook ingestion, background script execution, run history, and automatic reruns for failed runs — scaling it to tens of millions of automation runs per month at 99.95% success.

• •Designed RBAC-style permissioning and enterprise provisioning (SSO/SAML, SCIM, org-wide policies, audit logging) that unblocked adoption by large organizations and reduced admin-reported access issues ~30%.

• • Implemented two-way Sync and verified-data flows for the Connected Apps Platform, keeping shared datasets consistent across teams while cutting duplicate-data support tickets ~25%.

• •Drove a dedicated large-base performance effort, profiling hot paths and optimizing data access to improve load times on the largest bases ~40%.

• •Built full-stack features in React, TypeScript, and Node.js against internal REST APIs, and delivered internal tooling that let support teams diagnose failed automations and sync issues faster.

• •Partnered with product, design, and infrastructure teams; mentored junior engineers through code review and design guidance.

• •Built full-stack product features for Airtable's collaborative, relational app-building experience — linked records, views, comments, attachments, and sharing — as the product grew from a database tool into a workflow platform.

• • Contributed to the platform expansion into Blocks/Apps, the Scripting block, Automations, and Sync, turning Airtable into a connected operational data layer with JavaScript-powered custom workflows.

• •Developed frontend interfaces in JavaScript and React alongside backend REST API endpoints and data models supporting real-time collaborative editing.

• • Improved reliability and performance of core collaborative editing and workspace experiences, reducing sync related errors and improving perceived responsiveness for large shared bases.

• •Built and maintained developer-facing surfaces including the scripting environment and public API behavior (personal access tokens, OAuth, webhooks, scopes), supporting integrations with external SaaS tools.

• •Participated in on-call rotation, incident response, debugging, and cross-functional product planning within a fast scaling SaaS platform.