Deep Sci-Fi

A sci-fi writing workflow inspired by DeepMind's AI co-scientist.

This system uses competitive tournaments and AI debates to create science fiction grounded in research:

• Tournament competition generates multiple world-building scenarios
• Research-backed predictions create scientifically plausible futures through incremental steps
• Debate phases refine ideas through argumentation between different AI models
• Evolutionary refinement improves storylines, chapters, and prose through iterative competition

The result: rigorous research meets creative storytelling for scientifically-grounded sci-fi.

🏗️ Architecture

The system orchestrates a competitive tournament between AI models, with the Deep Sci-Fi Writer managing creative phases while Co-Scientist runs tournaments for idea generation and refinement. Research integration provides scientific accuracy through automated fact-gathering.

🚀 Quickstart

1. Clone the repository and activate a virtual environment:

git clone https://github.com/nerdsane/deep-sci-fi.git
cd deep-sci-fi
uv venv
source .venv/bin/activate  # On Windows: .venv\Scripts\activate

2. Install dependencies:

uv pip install -e .

3. Set up your .env file with API keys:

cp .env.example .env
# Edit .env with your API keys

4. Launch the system with LangGraph server:

uvx --refresh --from "langgraph-cli[inmem]" --with-editable . --python 3.11 langgraph dev --allow-blocking

Use this to open the Studio UI:

- 🚀 API: http://127.0.0.1:2024
- 🎨 Studio UI: https://smith.langchain.com/studio/?baseUrl=http://127.0.0.1:2024
- 📚 API Docs: http://127.0.0.1:2024/docs

📖 Deep Sci-Fi Writer

Creates scientifically-grounded science fiction through competitive AI tournaments and research integration.

Process

1. Storyline Competition: Multiple AI models compete to create storyline concepts
2. Research Integration: Automated fact-gathering for scientific accuracy
3. World Building: Develops plausible future scenarios and technologies
4. Chapter Development: Creates structured narrative with scientific explanations

Usage

Launch via LangGraph Studio, select "Deep Sci-Fi" graph, and enter your story concept. The system guides you through each stage with human-in-the-loop approval.

Configuration

Model templates (in src/co_scientist/configuration.py):

• Creative Template: Optimized for narrative tasks (Claude Opus + O3)
• Reasoning Template: Optimized for research tasks (O3 + GPT-4o)

🧬 Co-Scientist

Competitive tournament workflow that generates and evolves ideas through AI debates and quality ranking.

Core Features

• Tournament Competition: AI models compete to generate the best concepts
• Debate Phases: Different models argue for competing approaches
• Quality Ranking: Elo rating system tracks performance across competitions
• Evolutionary Refinement: Winning ideas are improved through iterative competition

Workflow Phases

1. Meta-Analysis: Generate multiple research/creative directions
2. Generation: Create detailed content for each direction
3. Reflection: Quality assessment and critique
4. Tournament: Head-to-head comparisons with debates
5. Evolution: Refine and improve winning concepts

Configuration

Templates automatically assign models per use case:

• Creative Template: For storylines, chapters (Claude Opus + Sonnet + O3)
• Reasoning Template: For research, analysis (O3 + GPT-4o + Sonnet)

Per-phase model override supported for advanced users.

📁 Project Structure

deep-sci-fi/
├── src/                          # Main source code
│   ├── co_scientist/            # Competitive tournament system
│   │   ├── phases/              # Tournament phases (debate, evolution, etc.)
│   │   ├── prompts/             # Phase-specific prompts
│   │   ├── utils/               # Model factory, LLM manager, output tools
│   │   └── configuration.py     # Model templates and settings
│   ├── deep_sci_fi/             # Main writing workflow
│   │   ├── deep_sci_fi_writer.py # Core orchestration logic
│   │   └── prompts.py           # Writing-specific prompts
│   └── open_deep_research/      # Research integration module
├── output/                      # Generated content (timestamped folders)
├── examples/                    # Usage examples and demos
└── langgraph.json              # LangGraph configuration

Output Organization

All generated content is saved to timestamped folders in output/:

output/YYYY-MM-DD_HH-MM-SS/
├── 00_01a_storyline_competition_summary.md    # Competition overview
├── 00_01b_storyline_competition_details.md    # Detailed results
├── 01_storyline_option_1_full.md              # Generated storylines
├── 02_world_scenario_full.md                  # World building
├── 03_first_chapter_full.md                   # Chapter content
├── elo_leaderboard.md                         # Quality rankings
└── tournament_*.md                            # Debate transcripts

Key Configuration Files

• src/co_scientist/configuration.py: Model templates, phase settings
• src/deep_sci_fi/deep_sci_fi_writer.py: Use case configurations
• langgraph.json: Workflow definitions and API settings

🤝 Contributing

1. Fork the repository
2. Create feature branch (git checkout -b feature/amazing-feature)
3. Commit changes (git commit -m 'Add amazing feature')
4. Push to branch (git push origin feature/amazing-feature)
5. Open Pull Request

📜 License

This project is licensed under the MIT License - see the LICENSE file for details.

🔗 Related Work

• Uses Open Deep Research for automated web research and scientific fact-gathering
• Inspired by DeepMind's AI co-scientist tournament-based hypothesis evaluation system