jarvis-research-os 1.0.0

JARVIS Research OS - AI-Powered Systematic Literature Review Assistant

JARVIS Research OS

AI-Powered Research Operating System for Systematic Literature Reviews

JARVIS Research OS is a local-first, AI-powered research assistant that helps researchers conduct systematic literature reviews with evidence grading, citation analysis, and contradiction detection.

Release Status

• Current release target: v1.0.0

✨ Features

Phase 1: Local-First Foundation

• Hybrid Search: Sentence Transformers + BM25 with RRF fusion
• Offline Mode: Network detection and graceful degradation
• Free APIs: arXiv, Crossref, Unpaywall integration

Phase 2: Differentiation Features

• Evidence Grading: CEBM evidence level classification (1a-5)
• Citation Analysis: Support/Contrast/Mention stance classification
• Contradiction Detection: Negation, antonym, and quantitative contradiction detection
• PRISMA Diagrams: PRISMA 2020 flow diagram generation (Mermaid/SVG)
• Paper Scoring: Comprehensive quality scoring with multiple signals
• Active Learning: Efficient screening with uncertainty sampling

Phase 3: Ecosystem

• MCP Hub: External search servers (PubMed/Semantic Scholar/OpenAlex)
• Browser Agent: Safe browser automation with security policy
• Skills System: Task-specific workflows via SKILL.md
• Multi-Agent Orchestrator: Parallel agents + approval flow
• Plugin System: Extensible architecture
• Zotero Integration: Reference management
• Export Formats: RIS, BibTeX, Markdown

🚀 Quick Start

Installation

# Clone repository
git clone https://github.com/kaneko-ai/jarvis-ml-pipeline.git
cd jarvis-ml-pipeline

# Install with uv (recommended)
uv sync

# Or with pip
pip install -e .

Basic Usage

from jarvis_core.evidence import grade_evidence
from jarvis_core.citation import extract_citation_contexts, classify_citation_stance
from jarvis_core.contradiction import Claim, ContradictionDetector

# Grade evidence level
grade = grade_evidence(
    title="A randomized controlled trial...",
    abstract="Methods: We conducted a double-blind RCT..."
)
print(f"Evidence Level: {grade.level.value} ({grade.level.description})")

# Analyze citations
contexts = extract_citation_contexts(text, paper_id="paper_A")
for ctx in contexts:
    stance = classify_citation_stance(ctx.get_full_context())
    print(f"Citation to {ctx.cited_paper_id}: {stance.stance.value}")

# Detect contradictions
detector = ContradictionDetector()
claim_a = Claim(claim_id="1", text="Drug X increases survival", paper_id="A")
claim_b = Claim(claim_id="2", text="Drug X decreases survival", paper_id="B")
result = detector.detect(claim_a, claim_b)
print(f"Contradiction: {result.is_contradictory}")

CLI Usage

# Search papers
jarvis search "machine learning cancer diagnosis"

# Run full pipeline
jarvis run --config pipeline.yaml

# Generate PRISMA diagram
jarvis prisma --output prisma_flow.svg

MCP Hub Example

# Register MCP servers and list tools
jarvis mcp list --config configs/mcp_config.json

# Invoke MCP tool
jarvis mcp invoke search_pubmed --params '{"query": "cancer immunotherapy"}'

Browser Agent Example

from jarvis_core.browser.subagent import BrowserSubagent
from jarvis_core.browser.schema import SecurityPolicy

policy = SecurityPolicy(url_allow_list=["pubmed.ncbi.nlm.nih.gov"])
agent = BrowserSubagent(security_policy=policy, headless=True)

Skills System Example

# List skills and show details
jarvis skills list
jarvis skills show MCP

📦 Core Modules

Module	Description
embeddings/	Sentence Transformers, BM25, Hybrid search
network/	Network detection, offline mode
sources/	arXiv, Crossref, Unpaywall clients
evidence/	CEBM evidence grading
citation/	Citation context and stance analysis
contradiction/	Claim normalization and contradiction detection
prisma/	PRISMA 2020 flow diagram generation
paper_scoring/	Paper quality scoring
active_learning/	Active learning for efficient screening

🧪 Testing

# Run all tests
uv run pytest

# Run specific module tests
uv run pytest tests/test_evidence_grading.py -v

# Run with coverage
uv run pytest --cov=jarvis_core

📖 Documentation

• API Reference
• User Guide
• Contributing

🔧 Configuration

Create config.yaml:

search:
  default_sources:
    - pubmed
    - arxiv
  max_results: 100

embeddings:
  model: all-MiniLM-L6-v2
  device: auto

evidence:
  use_llm: false
  ensemble_strategy: weighted_average

offline:
  enabled: true
  sync_on_connect: true

📄 License

MIT License - see LICENSE for details.

🙏 Acknowledgments

• Sentence Transformers
• PubMed
• arXiv
• PRISMA Statement