System Blueprint & Execution Graph

Ada Multi-Agent Architecture

A distributed, dual-corpus deep research engine designed for empirical citation grounding, adversarial gap arbitration, and resilient failover.

Backend Engine: Ada CoreMode: multi-agent-sync-executor
Qdrant ONNX (384d)
Jina 3-Tier Headless
Gemini 1M + Groq Failover
Execution Pipeline StagesSelect to inspect
01. Lead Planner AgentOrthogonal DAG Query Decomposition

Gemini 1.5/2.5 Flash • Structured JSON

Analyzes the user's research hypothesis and decomposes it into 3 to 7 orthogonal, non-overlapping investigation sub-questions. Formulates precise search queries targeting fundamental axioms rather than superficial keywords.

Contract Inputs
  • Raw Research Hypothesis
  • Depth Level (Quick vs Deep)
  • Domain Filter
Contract Outputs
  • Sub-question DAG array
  • Targeted search query matrix
  • Investigation categories
Resiliency & Fault Tolerance Guarantee

Regex-based trailing comma repair and automatic exponential backoff to handle free-tier API throttling.

Implementation Referenceengine/
def generate_research_plan(query: str, depth: str = "deep") -> List[Dict[str, Any]]:
    # Decomposes hypothesis into orthogonal research tracks
    prompt = f"Decompose into orthogonal research tracks: {query}"
    response = call_gemini_with_backoff(prompt)
    return parse_json_safely(response)