Access the live web deployments of the SLR Magic platform components directly in your browser:
| Application / Module | Live Deployment URL | Purpose |
|---|---|---|
| 🏠 SLR Magic Landing Page | https://iwas108.github.io/SLR-Magic/ | Platform overview, visual gallery & system documentation |
| 👥 Inter-Rater SPA | https://iwas108.github.io/SLR-Magic/inter-rater/dist/ | Offline double-blind human reviewer workspace |
| 📊 SLR Viewer SPA | https://iwas108.github.io/SLR-Magic/slr-viewer/dist/ | Interactive PRISMA 2020 flowchart & ECharts visualizer |
SLR Magic is an advanced, local-first platform designed to accelerate, calibrate, and standardize Systematic Literature Reviews (SLRs) for modern scientific research. By integrating stateful Large Language Model (LLM) automation with strict human-in-the-loop double-blind calibration, SLR Magic automates high-volume research tasks—such as multi-stage abstract screening, full-text validation, quality assessment, and structured taxonomy extraction—without compromising methodological rigor or data privacy.
┌──────────────────────────────────────────────────────────────────────────────────┐
│ SLR MAGIC PLATFORM ARCHITECTURE │
├───────────────────────────────────┬──────────────────────────────────────────────┤
│ 🚀 SLR IDE (slr-ide/) │ 👥 INTER-RATER SPA (inter-rater/) │
│ - Next.js 15 + SQLite Desktop │ - Standalone Offline React 19 SPA │
│ - 4-Stage Gemini LLM Pipeline │ - Blinded Human Reviewer Isolation │
│ - Turbovec Semantic Vector RPC │ - Cohen's Kappa (κ) Agreement Engine │
├───────────────────────────────────┼──────────────────────────────────────────────┤
│ 📊 SLR VIEWER (slr-viewer/) │ 📑 APP SCRIPT (app-script/) │
│ - Read-Only Snapshot Visualizer │ - Google Sheets FAIR Database Endpoint │
│ - Interactive PRISMA 2020 Canvas │ - Zero-OAuth Permission Security Model │
│ - 17 Scientific ECharts Panels │ - Automated Tabular Ingestion & Sync │
└───────────────────────────────────┴──────────────────────────────────────────────┘
Import paper collections from major bibliographic databases (Scopus, Web of Science, PubMed, IEEE Xplore) with dynamic column mapping and automated deduplication.
| Ingestion Hub Workspace | Raw Metadata Ingestion |
|---|---|
 |
 |
| Figure 1.1: Interactive CSV schema mapper with real-time duplicate preview. | Figure 1.2: Title similarity matching and DOI conflict resolution. |
Local PDF acquisition via multi-tier caching (MD5, DOI, title similarity, OCR page-1 matching) and headless scraper integration.
| Automated PDF Acquisition | Stealth Web Scraper Engine |
|---|---|
 |
 |
| Figure 2.1: Local PDF repository caching and cloud rclone sync. | Figure 2.2: Automated headless browser web scraper worker. |
Chained LLM execution powered by the Google Gemini Interactions API with stateful multi-turn context and cryptographic AES-256 API key encryption.
| Sequential LLM Operations | LLM Execution & Audit Log |
|---|---|
 |
 |
| Figure 3.1: 4-Stage screening (Fast Filter, Gatekeeper, Scientist, Miner). | Figure 3.2: Real-time execution monitor, token counter, and spend audit log. |
Blind human reviewers evaluate assigned paper pools independently to calculate statistical agreement before executing large-scale automated runs.
| Pre-Calibration Sandbox | Blinded Human Reviewer SPA |
|---|---|
 |
 |
| Figure 4.1: Pre-calibration agreement metrics and prompt tuning. | Figure 4.2: Standalone blinded human reviewer workspace. |
| Discrepancy Matrix & Cohen's Kappa | Pool Calibration Adjudication |
|---|---|
 |
 |
| Figure 4.3: Inter-rater agreement matrix and QA score alignment. | Figure 4.4: Adjudication panel resolving rater/AI discrepancies. |
Generate publication-ready 2D PRISMA 2020 flowcharts alongside 17 interactive ECharts analytics panels.
| Interactive PRISMA 2020 Canvas | Scientific Rigor & ECharts Analytics |
|---|---|
 |
 |
| Figure 5.1: 2D interactive PRISMA flowchart with SVG export. | Figure 5.2: Multi-level Sankey diagrams, radar charts, and spend grids. |
Export structured research datasets into FAIR-compliant .csv tables, .slr-viewer offline snapshots, or cloud-synced Google Sheets.
| FAIR Data Tabular Exporter | Google Sheets Endpoint |
|---|---|
 |
 |
| Figure 6.1: 100% complete CSV exporter with inline logic traces. | Figure 6.2: Synchronized FAIR database endpoint for collaborative review. |
The SLR Magic workspace consists of four modular applications:
1. 🚀 SLR IDE (slr-ide/)
- Role: Main desktop IDE & execution pipeline container.
- Tech Stack: Next.js 15 (App Router), SQLite (
slr.db), Python 3.10 Daemon,google-genai, Rclone. - Key Features: CSV Ingestion Hub, Smart PDF Cache Matcher, 4-Stage Gemini Screening Queue, Encrypted Key Vault, Turbovec RPC Vector Daemon, Pre-Calibration Sandbox, Wide Cohort Exporter.
- Role: Blinded human review & calibration web tool.
- Tech Stack: React 19, Vite 8, Tailwind CSS 4.
-
Key Features: Blinded evaluation mode, Cohen's Kappa (
$\kappa$ ) calculator, offlinelocalStoragepersistence, native.slrJSON schema file import/export.
3. 📊 SLR Viewer (slr-viewer/) • Live SPA
- Role: Standalone offline dashboard for dataset snapshot analysis.
- Tech Stack: React 19, Vite 8, Dexie.js (IndexedDB), Apache ECharts 6.
- Key Features: Offline
.slr-viewersnapshot import, interactive 2D PRISMA 2020 canvas, 17 scientific charts (Sankey, Radar, Stacked Bar), per-stage LLM spend grid.
- Role: FAIR-compliant database endpoint for Google Sheets.
- Tech Stack: Google Apps Script, Google Sheets, ECharts UI.
- Key Features: Direct CSV dataset ingestion, embedded sheet dialog visualizations, zero Google OAuth app permissions for maximum privacy.
SLR Magic enforces strict statistical validation targets defined in methodology.md:
| Screening Stage | Focus Area | Validation Target | Target Metric |
|---|---|---|---|
| Stage 1 (Fast Filter) | Title & Abstract Screening |
|
Zero false negatives |
| Stage 2 (Gatekeeper) | Full-Text Inclusion Verification |
|
Methodological soundness |
| Stage 3 (Scientist) | Quality Assessment (QA) | Ordinal tolerance |
|
| Stage 4 (Miner) | Structured Data Extraction | 0 missing JSON keys |
- Zero Database Leakage: SQLite files (
slr.db), downloaded PDFs, and local vector indexes are excluded from version control via strict.gitignorerules. - Encrypted Vault: LLM credentials are encrypted with AES-256-GCM + PBKDF2 inside local SQLite.
- Offline Capable: All core SPAs (
inter-rater/,slr-viewer/) run entirely offline in client browsers without sending data to external tracking servers.
cd slr-ide
npm install
# Setup Python environment for scrapers & Turbovec daemon
python -m venv python_engine/venv
source python_engine/venv/bin/activate # Windows: python_engine\venv\Scripts\activate
pip install -r python_engine/requirements.txt
# Start Next.js desktop engine
npm run devOpen http://localhost:3000 in your browser.
2. Launching Inter-Rater SPA (Or access Live Web Version)
cd inter-rater
npm install
npm run devOpen http://localhost:3001 in your browser.
3. Launching SLR Viewer (Or access Live Web Version)
cd slr-viewer
npm install
npm run devOpen http://localhost:3002 in your browser.
- 📐 System Architecture Blueprint (
architecture.md) - 🎯 Methodology & Calibration Standards (
methodology.md) - 📜 System Improvements History (
improvements.md) - 📘 SLR IDE Documentation (
slr-ide/README.md) - 📘 Inter-Rater SPA Documentation (
inter-rater/README.md) - 📘 SLR Viewer Documentation (
slr-viewer/README.md) - 📘 App Script Documentation (
app-script/README.md)
Distributed under the MIT License. See LICENSE for complete details.