Papers in. Ideas out.

ScholarScout reads 250M+ academic papers from 8 databases and generates actionable ideas
tailored to your goal: thesis, hackathon, SaaS product, literature review, or your next feature.

Quick Start · Four Modes · Features · Documentation · Live Demo · Changelog

---

Quick Start

git clone https://github.com/neej4/ScholarScout.git
cd ScholarScout
pip install -r requirements.txt
python preview_server.py

Open http://localhost:5050 — the setup wizard walks you through in 30 seconds.

Need an LLM? Pick one:

Provider	Cost	Speed	Setup
Gemini	Free (15 req/min)	Fast	Get key
Groq	Free tier	Very fast	Get key
Ollama	Free (local)	GPU-dependent	Download
Custom	Any	Any	Your local proxy (LM Studio, 9router)
OpenRouter	Pay-per-token	Varies	Get key
OpenAI	Pay-per-token	Fast	Get key

---

Four Modes

Same papers, four different lenses:

Mode	You ask	You get
Academic	"What can I research?"	Thesis topics, methodology, key papers, novelty check
Product	"What can I build?"	MVP features, tech stack, revenue model, competitors
Develop	"What can I add to my project?"	Features, integrations, optimizations grounded in your codebase
Review	"What's the state of the field?"	Thematic clusters, synthesis per cluster, gaps, open questions, reading list

Develop mode treats your project description as a hard constraint — every idea must be directly applicable.

Review mode doesn't generate ideas. It organizes and synthesizes existing papers into a literature review skeleton.

---

Features

Activity Center (v1.5.3)

• Owl Chase pixel art game while pipeline runs (papers spawn as dots you catch)
• Live graph showing papers grouped by category or cluster
• LLM Chat tab narrating what the AI is doing
• Adaptive phase list (5 phases default, 6 phases review)

Intelligence

• Trend analysis with confidence scoring
• Evidence Pack per generated idea: source papers, evidence claims, grounding score, and audit flags
• Anti-hallucination: P-number grounding (LLM citations are validated against fetched papers)
• Novelty check via semantic similarity (Gemini embeddings) or Jaccard fallback
• Quality scoring 1-10, low-quality filtered
• Deep dive: outline, methodology, datasets, timeline, tools, references
• Paper freshness: least-used papers prioritized, auto-widens date range when exhausted

Data

• 8 sources: arXiv, OpenAlex, Semantic Scholar, PubMed, Crossref, DOAJ, Scopus, DBLP
• Smart source routing per category (medicine → PubMed+Scopus, CS → arXiv+DBLP)
• 80+ categories across 10 disciplines
• Cache-aware with expiry (7 days configurable)
• Citation-based sorting

Personalization

• 18+ skill profiles (Academic, Product, Develop, Review)
• File upload (.pdf/.txt/.md/.json) as extra context
• Approach filter: Computational, Experimental, Clinical, Theoretical
• Onboarding wizard in 3 steps

Dashboard

• Real-time SSE streaming
• Search, filter, bookmark, export all ideas to Markdown
• Session recovery restores ideas plus cached deep dives/implementation scouting
• Evidence badges: Grounded, Partial, or Needs Review
• Session history (last 20 runs, review + default)
• Toast notifications (no browser alerts)
• Keyboard shortcuts

---

Project Structure

ScholarScout/
├── preview_server.py           # Entry point
├── run_pipeline.py             # CLI pipeline runner
├── config.example.yaml         # Config template (copy to config.yaml)
├── src/
│   ├── core/
│   │   ├── orchestrator.py     # Pipeline controller (default + review)
│   │   ├── analyzer.py         # Trend analysis
│   │   ├── generator.py        # 4-mode idea generation
│   │   ├── clusterer.py        # Paper clustering (review mode)
│   │   ├── synthesizer.py      # Literature synthesis (review mode)
│   │   ├── deep_dive.py        # Deep dive analysis
│   │   ├── novelty_checker.py  # Novelty scoring
│   │   ├── llm.py              # Multi-provider LLM client (6 providers)
│   │   ├── config.py           # Configuration + thresholds
│   │   ├── models.py           # Dataclasses
│   │   └── fetchers/           # 8 source fetchers
│   └── web/
│       ├── routes/             # Flask blueprints
│       ├── templates/          # Dashboard HTML
│       └── static/             # JS, sprites, owl game
├── skills/                     # ACADEMIC/ PRODUCT/ DEVELOP/ REVIEW/
├── tests/                      # 90+ automated tests
└── data/                       # Cache, snapshots, history (gitignored)

---

CLI Usage

# Academic mode
SCOUT_GOAL="THESIS" SCOUT_CATEGORIES="cs.AI,cs.CL" python run_pipeline.py

# Product mode
SCOUT_GOAL="HACKATHON" python run_pipeline.py

# Develop mode
SCOUT_GOAL="FEATURE" SCOUT_CONTEXT="Flask app with LLM integration" python run_pipeline.py

# Review mode
SCOUT_GOAL="SYNTHESIS" SCOUT_CONTEXT="federated learning for healthcare" python run_pipeline.py

---

Testing

pip install -e ".[dev]"
pytest tests/ --ignore=tests/integration    # Unit tests
npm test                                     # JavaScript tests

---

Contributing

See CONTRIBUTING.md. High-impact areas:

• New fetchers: implement BaseFetcher (1 file, ~150 LOC)
• New skill profiles: add markdown to skills/
• Prompt improvements: generator.py, analyzer.py, synthesizer.py
• New categories: update KEYWORD_SEEDS + fetcher mappings

---

Support

If ScholarScout saved you time, consider supporting:

• Ko-fi (International)
• Saweria (Indonesia)
• Star this repo

---

License

MIT — see LICENSE.