KYKirlos Yousef
All work

Product engineering · AI workspace

Spookling — AI Workspace

An AI workspace spanning native chat, calendar tools, browser-based coding, collaboration, search, and connected services.

Role
Product Engineer — AI Applications
Period
08/2025 — Present
Platforms
iOS · Web · Backend
Spookling app icon showing a friendly ghost in a witch hat on purple
Contribution

Built across iOS and web, then expanded into AWS backend services for chat, search, integrations, history, booking, and billing.

3 surfaces

iOS, web, and backend

Real-time

streamed AI and workspace chat

Product problem

Spookling brings always-available AI into chat, calendar workflows, coding, collaboration, and connected services. The work required more than a chat interface: native actions, progressive responses, workspace communication, search, external connections, and backend product flows had to behave as one dependable system.

Engineering approach

On iOS, SSE delivers partial model responses and typed tool boundaries protect native EventKit actions. The web product extends those workflows into browser-based workspaces and coding. Backend contributions added workspace and direct-message APIs, conversation files, OpenSearch-powered search, OAuth connections, assistant history, public booking, and billing fixes on AWS services.

Scope and evidence

The broader Spookling role began in August 2025. The documented backend contribution window covers July through September 2026. This case study describes implemented application features and infrastructure; it does not claim responsibility for model training or the entire product platform.

Engineering challenges

  1. Keep streamed model output responsive while maintaining predictable state across native and web clients.
  2. Translate evolving tool-call payloads into safe native actions and reliable third-party integrations.
  3. Add searchable workspace communication without exposing conversations across access boundaries.

Key decisions

  1. Used SSE for progressive AI responses and mapped tool calls through typed boundaries before EventKit actions.
  2. Built workspace and direct-message APIs on AppSync, Lambda, and DynamoDB with threads, mentions, and file indexes.
  3. Implemented OpenSearch Serverless indexing, highlights, and conversation-level access control.
  4. Added Buffer and Google connection flows, Composio tools, assistant history, booking APIs, and billing fixes.

Next case study

Sellou