Systematic Literature Review Multi-Agent Pipeline Architecture

Role: Principal Research Workflow Architect specializing in autonomous academic discovery pipelines.

Context

• Target Research Domain: {{research_domain}}
• Source Bibliographic Databases: {{target_databases}}
• Evidence Screening Criteria: {{quality_screening_criteria}}
• Human Oversight Checkpoint: {{human_review_stage}}
• Downstream Synthesis Deliverable: {{output_synthesis_format}}
• Institutional Data Governance Rules: {{governance_policy}}

Task

Design a comprehensive multi-agent workflow framework that automates the ingestion, filtering, deduplication, structured extraction, and synthesis of peer-reviewed literature within {{research_domain}}, ensuring methodological rigor and transparent state tracking across all agent transitions.

Method

1. Define agent node specializations (Ingestion Agent, Screening Agent, Extraction Agent, Synthesis Agent) and their respective context window budgets.
2. Construct the input ingestion schema to handle diverse querying syntax across {{target_databases}}.
3. Formulate the deterministic filtering logic and chain-of-thought screening rules using {{quality_screening_criteria}}.
4. Design the state persistence mechanism to preserve bibliographic metadata, inclusion rationale, and rejection logs.
5. Specify the structured data extraction schema for quantitative and qualitative findings.
6. Embed the mandatory human validation protocol at the {{human_review_stage}} to adjudicate borderline cases.
7. Establish the synthesis assembly chain to generate the final {{output_synthesis_format}}.
8. Define error handling, fallback retries, and rate-limit mitigation strategies across all agent execution stages.

Constraints

• MUST implement deterministic state schemas with zero ambient context bleed between pipeline stages.
• MUST NOT proceed past screening without logging explicit inclusion/exclusion justifications mapped to {{quality_screening_criteria}}.
• All agent handoffs MUST strictly comply with {{governance_policy}}.
• Architecture must include explicit fallback loops for parsing failures or API timeouts.

Output format

Provide the architectural framework across four structured sections:

1. Pipeline Node Architecture (table detailing agent roles, inputs, system prompts, outputs, and model configurations)
2. State Flow & Orchestration Protocol (step-by-step state transition graph, context boundaries, and retry policies)
3. Human-in-the-Loop Governance Specification (rubric and UI contract for {{human_review_stage}})
4. Verification & Audit Schema (telemetry tracking, inclusion matrices, and PRISMA compliance log formats)

Self-review

• Ensure every agent node has unambiguous input/output contracts.
• Verify that {{quality_screening_criteria}} is enforceable without human ambiguity prior to {{human_review_stage}}.
• Confirm that data governance constraints prevent hallucinated citation propagation.