Idea Catalyst
Find cross-domain research inspiration by decomposing problems, searching other fields, and synthesizing transferable insights.
The breakthrough for your research probably exists in a field you've never read — biologists solved the problem in 1998, or physicists have a framework that maps perfectly. But who has time to read papers in 15 other disciplines?
Who it's for: researchers stuck on hard problems who need fresh angles, PhD students looking for novel approaches, interdisciplinary scientists connecting disparate fields, R&D teams seeking innovation through cross-pollination, anyone who suspects the answer to their problem already exists somewhere else
Example
"Find cross-domain inspiration for our protein folding optimization problem" → Problem decomposed into 5 sub-questions, literature searched across biology, materials science, and operations research — 8 transferable insights ranked by applicability, with source papers and adaptation strategies
CLAUDE.md Template
New here? 3-minute setup guide → | Already set up? Copy the template below.
npx ccplaybooks add idea-catalyst# Idea Catalyst — Cross-Domain Research Inspiration
Find transferable insights from other scientific domains to spark novel approaches for your research problem. Based on the [Idea Catalyst](https://github.com/pkargupta/idea_catalyst) framework (Kargupta et al., 2025).
## When to Use
- You have a research problem and want inspiration from outside your field
- You're stuck on a challenge and want to explore cross-domain analogies
- You want to systematically discover how concepts from other fields could apply to your work
- You're writing a paper and need to identify novel cross-disciplinary connections
## How It WorksGet new playbooks like this one
One email a week with new Claude Code workflows. Free, like everything here.
No spam. Unsubscribe anytime.
What This Does
Systematically finds transferable insights from other scientific domains for your research problem. Based on the Idea Catalyst framework (Kargupta et al., 2025), it decomposes your problem into questions, searches your domain and cross-domain literature via Semantic Scholar, then integrates and ranks inspirations by transferability.
Quick Start
Step 1: Clone the Repository
git clone https://github.com/pkargupta/idea_catalyst.git
cd idea_catalyst
pip install -r requirements.txt
Step 2: Download the Template
Click Download above to get the CLAUDE.md file and place it in the idea_catalyst/ directory.
Step 3: Start Working
claude
Say: "Find cross-domain inspiration for my research problem: [describe your challenge]"
The Four-Stage Pipeline
| Stage | What Happens |
|---|---|
| 1. Decompose | Break research problem into targeted questions (technical, conceptual, methodological) |
| 2. Target Search | Search your domain's literature for current landscape and gaps |
| 3. Cross-Domain Search | Reformulate questions as domain-agnostic, search other fields |
| 4. Integrate & Rank | Synthesize cross-domain findings into ranked inspiration ideas |
Prerequisites
- Python 3.10+
- Semantic Scholar API key (free)
- GPU recommended for local LLM inference (uses vLLM)
- Dependencies:
torch,transformers,vllm,spacy,pandas,scikit-learn
Key Options
| Flag | Description |
|---|---|
--model_name |
LLM for decomposition and synthesis (default: Qwen3-14B) |
--max_papers_per_query |
Papers to retrieve per search query |
--min_rel_threshold |
Minimum relevance score to keep (higher = fewer, better results) |
--skip_if_exists |
Resume interrupted batch runs |
Using Without the Pipeline
You can use the Idea Catalyst thinking framework directly with Claude — no setup needed:
- Describe your problem — be specific about the research challenge
- Decompose — "Break this into 3-5 targeted research questions"
- Cross-domain search — "What fields outside [your domain] have solved similar problems?"
- Synthesize — "How could [concept from another field] adapt to my problem?"
Tips
- Start with a clear, specific problem statement — vague problems yield vague inspirations
- Try multiple target domains — the best insights often come from unexpected fields
- The
min_rel_thresholdcontrols quality vs. quantity: higher = fewer but more relevant - The dataset is derived from CHIMERA cross-domain inspiration relations
Example Prompts
"Find cross-domain inspiration for improving transformer efficiency in NLP"
"What can drug discovery learn from supply chain optimization?"
"Decompose my research problem into cross-domain searchable questions"
"Search biology and physics for analogies to my distributed systems challenge"
"Rank these cross-domain ideas by transferability to my problem"