Responsibilities
- Lead major components of one or both 2025 technical initiatives: advanced AI agent systems with real-time interface or real-time multiplayer document collaboration
- Develop agent workflows using LangGraph for stateful, long-running reasoning loops
- Design and implement tool orchestration, memory management, retrieval mechanisms, and error resilience in agent execution
- Enhance semantic and hybrid search performance for accuracy and speed
- Build real-time agent chat interfaces using websockets, PartyKit, and Vercel's streaming protocol
- Implement collaborative features including comments, tagging, mentions, and notifications
- Develop frontend interfaces using React, Next.js, TypeScript, Tailwind, shadcn, and Framer Motion
- Construct MCP Servers in FastAPI to support agent execution and secure data retrieval
- Manage streaming protocols, authentication flows, and context filtering based on user permissions
- Implement real-time document synchronization using Y.js and CRDT algorithms
- Develop rich-text editing capabilities using TipTap and ProseMirror
- Integrate version control with change tracking, diff visualization, and recovery from historical states
- Enable real-time collaboration features such as comments, tagging, and notifications
- Enforce granular, permission-aware access controls at both document and content block levels
- Ensure document updates are immediately available in search results
- Optimize retrieval-augmented generation (RAG) pipelines for incremental indexing and access-controlled filtering
- Improve document ingestion systems using LlamaCloud, Unstructured, and Docling
- Expand and stabilize integrations with Google Workspace, Microsoft 365, and various database systems
Work Arrangement
Remote (Worldwide)
Projects for You
You will take ownership of significant parts of one or both of the two major 2025 engineering initiatives: Deep Agentic RAG with Realtime Agent UI or Multiplayer Document Editing.
Project 1: Deep Agentic RAG and Realtime Agent UI
- We are developing a high-performance ReAct-style agent with extended reasoning capabilities (Deep Agents), fully integrated into the workspace. This includes a powerful agent engine, stateful workflows via LangGraph, a collaborative real-time chat UI, a developer-focused interface, and MCP Servers enabling external tools to run agents and retrieve context.
- Agent Engine Requirements: Build agent loops with LangGraph, implement tool orchestration, memory, retrieval, and error handling, and optimize semantic and hybrid search.
- UI and Realtime Requirements: Develop real-time agent chat using websockets, PartyKit, and Vercel streaming; implement comments, tagging, mentions, and notifications; use React, Next.js, TypeScript, Tailwind, shadcn, and Framer for frontend development.
- MCP Integration Requirements: Build FastAPI-based MCP Servers for agent execution and data retrieval, and manage streaming, authentication, and permission-aware context.
- Non-Goals: This project does not include an agent marketplace, plugin ecosystem, or integration with non-permissioned data sources.
- Prior Art: Inspired by LangGraph Deep Agents (architecture), Anthropic MCP (server design), Cursor agent UI (UX patterns), and Liveblocks/PartyKit (multiplayer communication).
Project 2: Multiplayer Document Editing
- We are transforming documents into real-time collaborative workspaces with deep AI integration, enabling multiplayer synchronization, versioning, commenting, tagging, searchability, and direct links to ingestion and connector pipelines.
- Editor Architecture Requirements: Implement real-time sync using Y.js and CRDTs, build rich-text editing with TipTap and ProseMirror, and add version control, diff tracking, and historical recovery.
- Collaboration Layer Requirements: Support real-time comments, tagging, notifications, and enforce permission-aware access at document and block levels.
- Search and RAG Integration Requirements: Make document changes instantly searchable and optimize RAG pipelines for incremental updates and access filtering.
- Connector and Ingestion Requirements: Improve ingestion using LlamaCloud, Unstructured, and Docling; strengthen integrations with Google Workspace, Microsoft 365, and databases.
- Non-Goals: This effort does not aim to build a full office suite, support binary file editing, or enable offline editing.
- Prior Art: Informed by Google Docs (collaborative editing), Notion (commenting and tagging), Linear (collaboration UI), Yjs demos (CRDT sync), and VS Code Live Share (multiplayer reliability).
Other
- Applications must be submitted by 8pm Pacific Time on Monday, December 8, 2025.
- Interviews will be completed and decisions communicated by December 12.
