Systematic Literature Review Agentic Pipeline Framework
Design a robust multi-agent workflow framework for automated academic literature retrieval, screening, and thematic synthesis.
Use this template when designing or upgrading an automated pipeline that searches, screens, extracts, and summarizes academic literature across distributed repositories. It establishes clear chain dependencies, human-in-the-loop validation checkpoints, and hallucination guardrails for research teams.
Role: Principal Research Informatics Architect with twenty years of experience designing multi-agent data pipelines for higher education.
Context
- Host Institution: {{institution_name}}
- Research Discipline: {{research_domain}}
- Target Bibliographic Repositories: {{source_repositories}}
- Methodological Screening Bar: {{quality_screening_threshold}}
- Formatting Standard: {{citation_style}}
- Oversight Checkpoint: {{human_review_stage}}
Task
Construct a comprehensive operational framework for an end-to-end multi-agent workflow chain that ingests raw search queries, extracts academic manuscripts from {{source_repositories}}, executes automated methodology screening for {{research_domain}}, and prepares verified synthesis drafts.
Method
- Define ingestion nodes responsible for query expansion, cross-database querying across {{source_repositories}}, and deduplication of candidate manuscripts.
- Detail the schema extraction agent that parses full texts and metadata according to {{citation_style}} conventions.
- Formulate the triage agent's decision tree enforcing {{quality_screening_threshold}} based on sample sizes, peer-review status, and statistical rigor.
- Design the synthesis reasoning agent that groups findings into thematic clusters relevant to {{research_domain}}.
- Specify inter-agent message payloads, state schemas, and deterministic fallback protocols for malformed inputs.
- Architect the boundary conditions triggering {{human_review_stage}} before any synthesis is finalized for {{institution_name}}.
- Establish automated provenance tracing linking every summary statement back to original source paper paragraph identifiers.
Constraints
- MUST include explicit state-transition diagrams represented in structured markdown tables.
- MUST NOT permit ungrounded claims; every synthesized point must cite extracted paper records.
- Screening criteria must reflect specific quality thresholds defined by {{quality_screening_threshold}}.
- Provide concrete error-handling mechanisms for rate limits and paywalled documents.
Output format
- Workflow Architecture Map: sequential agent nodes, inputs, and outputs
- Screening & Triage Protocol: automated scoring logic based on {{quality_screening_threshold}}
- Human-Agent Interface Specification: exact triggers and interface for {{human_review_stage}}
- Provenance & Verification Audit Rules: data validation rules in {{citation_style}}
Self-review
- Did I map every agent handoff with typed payload schemas?
- Are fallback routes defined for ambiguous or low-confidence extractions?
- Does the framework strictly enforce human oversight at {{human_review_stage}}?
Explicit role, a named task, and discrete steps the model can follow.
Background, inputs and variables the model needs before it starts.
Hard boundaries — what the model must and must not do.
A named, field-level shape for the response.
Ordered work items that force analysis before an answer.
Length and structure that travel across frontier models.
Signal density — instruction weight without padding.
Documented variables so the scaffold adapts to new inputs.
Quality bar, assumptions and behaviour when inputs are thin.
How much real usage the template has behind it.