agentSYSTEM_ID: agentic-audit-harness

Autonomous Agentic Evaluation & Guardrail Harness

LangGraph cyclic state machine with deterministic input guardrails, embedded SQLite memory, and real-time hallucination evaluation.

Autonomous Agentic Evaluation & Guardrail Harness // Audio BriefingSTUDIO MP3

0:00 / 0:45

VISUAL ARCHITECTURE // N8N WORKFLOW GRAPH

CLICK ANY NODE TO INSPECT LIVE BLUEPRINT

🌐 TRIGGER

User / Recruiter Prompt

Audit Query Intake

🛡 SAFETY GATE

Input Guardrail Node

Zero-Latency Scanner

🗄 PERSISTENCE / RAG

SQLite Factual Memory

data/arun_sys.db

⚡ COGNITIVE CORE

Gemini 2.5 Flash Node

Grounded Reasoning

🛡 SAFETY GATE

Evaluation & Citation Gate

Hallucination Scorer

🗄 PERSISTENCE / RAG

SQLite Session Persist

Audit Turn Logged

Gemini 2.5 Flash Node// Grounded Reasoning

⚡ COGNITIVE CORE

Executes LLM reasoning grounded strictly in retrieved SQLite facts with structured citations.

EXECUTION SPEC // PYTHON IMPLEMENTATION

def gemini_inference_node(state: AgentState) -> dict:
    response = client.models.generate_content(
        model="gemini-2.5-flash",
        contents=state["query"]
    )
    return {"raw_response": response.text}