← Work
LiveSr. Product Manager — AI & Innovation · Ting Works LLP · 2025-04present

Project ATOM

An intelligence layer for event ingestion, inference pipelines, semantic search, and role-based access. Measured in hours returned to the team, not demos given.

RAGMulti-agentRBACpgvectorVertex AI

Project ATOM is the central nervous system for Ting Works. It connects event streams from across the company, runs them through a controlled inference pipeline, and surfaces context to humans through five micro-systems: ZPublish, TCIS, Workflow Manager, Tingdom Gateway, and HR Intelligence. Role-based access is enforced at the retrieval layer, not as an afterthought.

The architectural bet is simple. Authorization is not a feature you bolt on. In a multi-agent system it is the architecture. Every retrieval is filtered by RBAC. Every inference call is audited. Every output carries provenance back to source events. The system is measured in hours returned to the team, not demos given.

The work itself looks unglamorous up close. Build the event bus. Write the prompt contracts. Train internal users on what to trust and what to verify. Ship incrementally so each module pays for the next.

// stack
ingest event bus · webhooks · supabase streams
retrieve pgvector · hybrid · role-filtered
infer vertex ai · llama · prompt contracts
govern rbac · audit · provenance
surface zpublish · tcis · workflow mgr · gateway · hr

Impact

  • 30–40% manual work removed across surveyed teams
  • First internal knowledge graph at Ting
  • RBAC and audit logs from day one. No retrofit.