PassedAI

Problem Solved

Enterprise knowledge bases decay silently. Documentation becomes outdated, contradictory, or concentrated in the hands of a few experts—creating operational risk that's invisible until something breaks. Employees waste time following stale procedures, get confused by conflicting guidance, or can't find answers to common questions.

PassedAI solves this by continuously auditing knowledge bases to detect four types of decay:

• Stale documents not updated in 6-24+ months
• Contradictions where two docs give opposite guidance
• Knowledge gaps from repeated unanswered questions
  
• Expert risk where critical docs have single owners

Features Used

• Elastic Agent Builder with 2 custom agents (Detect + Act) connected via A2A protocol
• 4 ES|QL tools for staleness scoring, gap detection, contradiction finding, and expert risk analysis
• ELSER v2 (semantic_text field type) for semantic search in contradiction detection
• Elasticsearch Serverless for zero-ops deployment
• Python orchestrator coordinating agents via JSON-RPC

What I Liked

1. ES|QL tools are powerful — Writing custom queries that agents can invoke autonomously makes data analysis feel native to the AI workflow.

2. A2A protocol enables multi-agent patterns — The Detect→Act handoff creates a clean separation between analysis and remediation.

Challenges

1. A2A message format discovery — Had to debug JSON-RPC payloads to find that Elastic uses kind: "text" instead of type: "text" in message parts.

2. Serverless restrictions — Index mappings can't include shard/replica settings; bulk ingestion with semantic_text requires longer timeouts (120s) due to ELSER inference time.

Built With

• a2a-protocol
• ai
• elastic
• elastic-agent-builder
• elasticsearch-serverless
• elser-v2
• es|ql
• flask
• python
• tailwind-css