Projects

A curated selection of projects that demonstrate technical depth, practical judgment, and end-to-end ownership.

Connected Knowledge Platform

A platform enabling Zendesk customers to ingest knowledge from third-party sources directly into their Help Center for AI-powered search and agent assistance.

Problem

Customers' knowledge lives scattered across dozens of external tools. Without this content in Zendesk, AI features like generative search and Agent Copilot operate on incomplete information, reducing answer quality and forcing agents to context-switch between systems.

Role

Technical lead — authored the RFC, designed the architecture, built the shared component library, and delivered the end-to-end integration across frontend and backend.

Approach

Designed a connector-agnostic UI framework with a registry-driven architecture so new connectors could be added with minimal code. Built a multi-step wizard (OAuth, folder selection, sync configuration), integrated with Paragon for third-party auth orchestration, and gated rollout per-connector using feature flags and an EAP framework.

Outcome

Shipped 8+ connectors from RFC to GA within ~3 months. The platform architecture allowed new connectors to be added in days rather than weeks. Enabled Zendesk's AI features to operate on a much broader knowledge base.

TypeScriptReactOAuth 2.0ParagonGraphQLFeature Flags

Quick Answers (Generative Search)

An AI-powered feature that generates direct answers to user search queries in Help Center, synthesized from existing knowledge base articles using RAG.

Problem

End users searching Help Center had to manually sift through article results to find answers. The existing search only returned links — causing frustration and increased ticket volume for common questions.

Role

Full-stack developer — owned the feature from backend prompt/endpoint work through to frontend rendering, formatting, accessibility, and rollout.

Approach

Built a custom Markdown rendering pipeline for generative answers with proper formatting, collapsible text, and accessibility support (VoiceOver, ARIA roles, contrast). Integrated into both Help Center (end-user) and the Agent Knowledge Panel.

Outcome

Reduced end-user effort to find answers in Help Center. Feature expanded from end-user search to Agent Copilot, becoming a core pillar of Zendesk's AI-assisted support strategy.

TypeScriptReactRuby on RailsRAGMarkdownARIA

Macros Admin Rewrite

A ground-up rewrite of the Zendesk macros management experience — modern Admin Center with rich text editor, AI-powered suggestions, and manual reordering.

Problem

The existing macros management interface was outdated, lacked a rich editor, didn't support macro reordering, and couldn't accommodate AI-powered suggestion features. Admins managing hundreds of macros needed a modern, performant tool.

Role

Primary frontend engineer delivering the new experience end-to-end.

Approach

Built a rich text editor with placeholder support, inline images, hyperlink handling, and RTL localization. Implemented AI macro suggestions with acceptance feedback loops. Added Sentry error tracking and configured route migrations.

Outcome

Delivered a fully modernized macros admin experience that unlocked AI suggestions for macro creation and enabled manual macro ordering — both highly requested by admins managing large macro libraries.

TypeScriptReactRich Text EditorSentryGraphQL

Agent Copilot Knowledge Integration

Integrating AI-powered knowledge features into the Agent workspace — suggested content, suggested actions, and voice copilot within the Knowledge Panel.

Problem

Agents handling tickets had to manually search for relevant articles. The Knowledge Panel needed to proactively surface AI-generated suggestions, but the existing architecture couldn't accommodate these new data sources or interaction patterns.

Role

Frontend engineer building the data layer and UI for multiple copilot features within the Knowledge Panel.

Approach

Built data fetching and rendering for Agent Copilot and Voice Copilot, implemented suggested actions with feedback submission, added contextual workspace filtering, and tracked content suggestion usage via Datadog metrics.

Outcome

Enabled agents to receive proactive, context-aware knowledge suggestions without leaving their workflow — feeding into Zendesk's broader Agent Copilot product offering with measurable usage tracking.

TypeScriptReactGraphQLDatadogZendesk Garden