Back to Agents
Category I

Category I: Systematic Review Agents

PRISMA 2020 pipeline automation with AI-assisted screening and RAG

Systematic Review agents automate the complete PRISMA 2020 literature review workflow. From multi-database paper retrieval to AI-assisted screening and vector database construction, they handle the technical complexity so researchers can focus on synthesis.

Core Principle

Automated but transparent — every decision is logged, every exclusion is justified, every step follows PRISMA 2020 guidelines

SR Pipeline Orchestrator

I0

Coordinate the 7-stage systematic review pipeline, manage stage transitions and validation

OpusHIGHNone

Trigger Keywords

systematic reviewPRISMAliterature review automationpipeline

Capabilities

  • Stage-by-stage pipeline coordination (7 stages)
  • Prerequisite validation before stage transitions
  • Project initialization and configuration management
  • Progress tracking and status reporting
  • PRISMA flow diagram generation

VS Process

Direct orchestration — no VS needed for pipeline coordination

Example

Input: "I want to conduct a systematic review on AI in education"
Output: Initializes project → Stage 1: Research domain setup → Validates scope → Stage 2: Query design → Continues through all 7 stages with checkpoints

Paper Retrieval Agent

I1

Fetch papers from multiple academic databases with parallel queries and deduplication

SonnetMEDIUM🔴 CP_DATABASE_SELECTIONNone

Trigger Keywords

fetch papersdatabase searchpaper retrievalSemantic ScholarOpenAlexarXiv

Capabilities

  • Multi-database parallel querying (Semantic Scholar, OpenAlex, arXiv)
  • Institutional database support (Scopus, Web of Science)
  • DOI-based and title-similarity deduplication
  • API key validation and rate limit management
  • PDF URL extraction and availability checking

VS Process

Database selection validated at checkpoint before execution

Example

Input: "Search for papers on AI chatbots in language learning from 2020-2024"
Output: 🔴 CP_DATABASE_SELECTION → User selects databases → Parallel fetch: Semantic Scholar (1,200 results) + OpenAlex (980) + arXiv (340) → Deduplicate: 1,890 unique papers

Screening Assistant

I2

AI-assisted PRISMA 6-dimension screening with configurable LLM providers

SonnetMEDIUM🔴 CP_SCREENING_CRITERIANone

Trigger Keywords

screeninginclusion criteriaexclusion criteriarelevance scoringPRISMA screening

Capabilities

  • AI-PRISMA 6-dimension relevance scoring
  • Configurable LLM providers (Groq, Claude, Ollama)
  • Title/abstract screening with threshold management
  • Exclusion reason tracking and categorization
  • Human validation sampling for inter-rater reliability

VS Process

Screening criteria validated at checkpoint before AI scoring begins

Example

Input: "Screen 1,890 papers for relevance to AI chatbot language learning"
Output: 🔴 CP_SCREENING_CRITERIA → Define inclusion/exclusion → Groq LLM scores papers → Auto-include: 245 (score>40) | Auto-exclude: 1,420 (score<20) | Manual review: 225 (borderline)

RAG Builder

I3

Construct vector databases from collected PDFs using local embeddings for zero-cost semantic search, with parallel document processing for large corpora (absorbed B5)

HaikuLOWNone

Trigger Keywords

RAGvector databaseembeddingsChromaDBsemantic searchparallel processing

Capabilities

  • PDF text extraction with PyMuPDF and OCR fallback
  • Local embedding generation (sentence-transformers, zero cost)
  • ChromaDB vector database construction
  • Configurable chunk size and overlap settings
  • Ingestion logging and quality validation
  • Parallel document processing for large-scale corpora (absorbed B5)

VS Process

Direct execution — validates PDF collection completeness before building

Example

Input: "Build RAG system from 180 downloaded PDFs"
Output: Extract text from 180 PDFs (parallel processing) → Generate embeddings (local, $0 cost) → Create ChromaDB index → 15,400 chunks indexed → Ready for semantic queries

Pipeline Coverage

Knowledge Repository (15K-20K papers, 50% threshold) and Systematic Review (50-300 papers, 90% threshold)

Integration with Other Categories

  • Category A (Foundation): Research question from A1 drives search strategy
  • Category B (Evidence): B1 literature review strategy informs query design
  • Category C (Design): C5 meta-analysis protocol feeds Stage 3 configuration
  • Category G (Communication): G2 Publication Specialist validates PRISMA checklist compliance at each stage

Checkpoint Information

I1 (Paper Retrieval) requires CP_DATABASE_SELECTION (🔴 REQUIRED) before fetching. I2 (Screening) requires CP_SCREENING_CRITERIA (🔴 REQUIRED) before AI scoring begins. These ensure researcher control over critical pipeline decisions.

Best Practices

  • Use all available databases: Maximize coverage by combining open access (3) and institutional (2) sources
  • Validate screening criteria: Have domain expert review inclusion/exclusion criteria before AI screening
  • Monitor PDF retrieval rates: Expect 50-60% success; supplement with manual retrieval for critical papers
  • Test RAG queries: Verify vector database quality with known-answer queries before full analysis

Auto-Trigger Examples

User Input: "I want to do a systematic review on AI in education"
Detected: Keywords: "systematic review" → Triggers I0 (Pipeline Orchestrator)
Execution: I0 initializes project → Stage 1 conversation → Validates research scope → Proceeds to Stage 2
User Input: "Fetch papers from Semantic Scholar and OpenAlex about chatbot learning"
Detected: Keywords: "fetch papers", "Semantic Scholar", "OpenAlex" → Triggers I1 (Paper Retrieval)
Execution: 🔴 CP_DATABASE_SELECTION → I1 validates APIs → Parallel fetch → Deduplication → Results CSV