Unified Research Impact Dashboard
Personal research portfolio + 147-school graduate advisor directory + AI-native talent search.
English | 中文
Features • Quick Start • Configuration • Architecture
| Route | Purpose |
|---|---|
/profile/:id |
Per-researcher dashboard (papers / repos / HF / persona / capability / poem / share cards) |
/leaderboard |
Total + young scholar + 6 direction leaderboards (top 10000 exact rank, then percentile) |
/recruit |
B2B AI talent search — recruiter pastes JD → LLM extracts criteria → DB filter → LLM rerank with reasoning |
/advisor |
Graduate advisor directory — 147 双一流 (985/211) universities, LLM-driven crawler for colleges + faculty stubs |
/users |
All ImpactHub profiles |
/docs |
System overview |
One profile that unifies your presence across Semantic Scholar, GitHub, and Hugging Face. Enter your Scholar ID — the system auto-discovers your linked GitHub and HF accounts.
- H-index auto-computation and CCF-A/B/C venue classification
- Identifies top scholars (h-index ≥ 50) and notable scholars (h-index ≥ 25) who cite your work
- LLM-powered honor tag enrichment — detects IEEE Fellow, ACM Fellow, 院士 among your citers
- Per-paper citation drill-down with context snippets
- Daily metric snapshots: citations, h-index, stars, forks, downloads, likes
- Interactive trend charts with 30/60/90/365-day windows
- Milestone system: automatic achievements when you hit thresholds (100 citations, 1K stars, etc.)
- Perplexity-powered web search to gauge your research visibility
- Heat level classification (hot / medium / cold) with source links
- LLM-generated researcher bio capturing your research identity
- Auto-generated research tags from your publication topics
- LLM picks one of 12 meme personas (GOAT / PI / WOLF / JUAN / KPI / MONK / NINJA / PROOF / SENSEI / VIRAL / BDFL / HYPE) with AI-generated illustrations
- 4-axis dimensions (output / ecosystem / seniority / collaboration) with continuous scores
- LLM identifies 1-4 research directions a scholar actually works in
- Per-direction role (originator / early_adopter / extender / follower) + weight + achievements + representative works
- Avoids the single-tag oversimplification
- Tree-shaped research trajectory: time on Y axis, research branches on X
- Xiaohongshu-style annual research poem (10-14 verses, data-driven)
- LLM + web-search career timeline (education + positions, with sources)
- Total / Young Scholar (first paper <10y) / 6 directions (LLM, CV, VLM, Systems, Theory, RL)
- Top 10000 shows exact rank; beyond that shows percentile only
- Free-text JD → LLM extracts structured criteria (direction, seniority, h-index threshold, must/nice keywords, honors)
- DB coarse filter → top 40 candidate dossiers → LLM rerank with
match_score,tier, fit reasoning, concerns, key works
- Seed: 147 双一流 universities (incl. all 985/211), authoritative source: 教育部 2022 第二轮双一流 list
- LLM-driven generic crawler — no per-school adapters; uses heuristic college/teacher extraction with JS-redirect follow + permissive SSL for old .edu.cn
- Three-layer pipeline: school directory → college list → advisor stubs (name + title + homepage)
- Research Basis Generator for NSFC, Changjiang, Wanren, and other Chinese grant types
- Tone-adaptive formatting: "potential + feasibility" for youth grants vs. "originality + leadership" for senior grants
- Paper selection UI with evidence preview (citation analysis + notable scholars + linked repos)
- Paper list export: Markdown, BibTeX, JSON
- Filter by year, CCF rank, citation count, first-author
- Comprehensive CV-style summary JSON
- Background scheduler refreshes all data every 6 hours
- Manual refresh on demand via API
- Python 3.11+
- Node.js 18+
- An OpenAI-compatible API key (for AI summary & buzz features)
cd backend
pip install -r requirements.txt
# Create .env from template
cp ../.env.example .env
# Edit .env and fill in your API key
python -m uvicorn app.main:app --host 0.0.0.0 --port 8001cd frontend
npm install
# Development
npm run dev
# Production build (served by backend)
npm run build# Serves frontend dist/ + proxies /api to backend
python serve.py 19487Open http://localhost:19487 and enter your Semantic Scholar ID to get started.
Copy .env.example to backend/.env:
| Variable | Description | Required |
|---|---|---|
LLM_API_BASE |
OpenAI-compatible API endpoint | Yes |
LLM_API_KEY |
API key for the LLM provider | Yes |
LLM_BUZZ_MODEL |
Model for buzz & summary generation (default: gpt-5) |
No |
OUTBOUND_PROXY |
HTTP proxy for outbound API calls | No |
GITHUB_TOKEN |
GitHub PAT for higher rate limits | No |
impacthub/
├── backend/
│ ├── app/
│ │ ├── main.py # FastAPI entry + static file serving
│ │ ├── config.py # Environment & constants
│ │ ├── models.py # SQLAlchemy ORM
│ │ ├── schemas.py # Pydantic schemas
│ │ ├── routers/
│ │ │ ├── profile.py # Profile CRUD & account linking
│ │ │ ├── stats.py # Aggregated statistics
│ │ │ ├── citations.py # Citation analysis & scholar classification
│ │ │ ├── growth.py # Growth snapshots & trends
│ │ │ ├── milestones.py # Achievement tracking
│ │ │ ├── buzz.py # Web presence monitoring
│ │ │ ├── ai_summary.py # LLM-generated bios & tags
│ │ │ ├── reports.py # Grant research basis generator
│ │ │ └── data.py # Export endpoints
│ │ ├── services/ # Business logic per domain
│ │ └── tasks/
│ │ └── scheduler.py # APScheduler (6h refresh cycle)
│ └── requirements.txt
├── frontend/
│ ├── src/
│ │ ├── pages/ # Setup, Profile, Milestone, Users
│ │ ├── components/ # Charts, cards, modals, exporters
│ │ └── lib/ # API client, utils, venue data
│ └── package.json
└── serve.py # Simple dev proxy server
Tech Stack: FastAPI + SQLAlchemy + aiosqlite | React 19 + Tailwind CSS 4 + Recharts | Semantic Scholar + GitHub + Hugging Face APIs
MIT
