Lean In

UK politics for the modern audience.

Built at a the Claude Hackathon at Northumbria University. Designed to cut through political nonsense so voters can make more informed decisions.

---

What it does

Open the map. Pick a nation. Type any policy topic — renters rights, NHS waiting times, tuition fees. Every major party's position appears as a card, each claim a direct quote from that party's 2024 General Election manifesto with a page number you can verify.

Search an MP's name instead and you get their profile: role, constituency, voting record, and where their party stands.

---

Stack

Layer	Tech
Backend	Python 3.11, FastAPI, Anthropic SDK, pydantic
Frontend	React 18 + Vite, plain CSS, Leaflet
LLM	claude-sonnet-4-6 with prompt caching
Data	ONS GeoJSON, official 2024 manifesto PDFs, Democracy Club

No database. No vector store. No embeddings. Manifestos sit in the Claude system prompt with cache_control: ephemeral.

---

Quick start

Prerequisites: Python 3.11+, Node 20+, an Anthropic API key.

git clone https://github.com/Abdulla-AlBassam/lean-in
cd lean-in

Backend (Terminal 1):

cd backend
python3.11 -m venv .venv

# Mac/Linux
source .venv/bin/activate

# Windows
.\.venv\Scripts\Activate.ps1

pip install -r requirements.txt
cp .env.example .env          # then add your key: ANTHROPIC_API_KEY_1=sk-ant-...
uvicorn app.main:app --reload --port 8000

Frontend (Terminal 2):

cd frontend
npm install
npm run dev                   # → http://localhost:5173

Verify: curl http://localhost:8000/api/health → {"status":"ok"}

Environment variables

Variable	Required	Description
ANTHROPIC_API_KEY_1	Yes	Anthropic API key. Add more as _2, _3 — the backend round-robins across all of them.
DEMO_MODE	No	Set to true to serve all responses from backend/demo_cache/ without hitting the API.

---

API

Endpoint	Description
GET /api/health	Liveness check
GET /api/search?q=&nation=	Topic search or person search. nation = UK | ENG | SCO | WAL
GET /api/person/{id}	Full MP profile + timeline
GET /api/axes	Political axis definitions with party positions

Full request/response shapes in contracts.md.

---

Bias defence

Every quote shown in the UI is a verbatim substring of the manifesto file in this repo. The topic classifier is deterministic keyword matching — no LLM involved. The LLM only summarises and extracts; it cannot invent positions because it only sees the text we pass it. If a party's manifesto is silent on a topic, the response says so plainly.

---

Project structure

/
├── backend/
│   ├── app/
│   │   ├── main.py        routes + person detection
│   │   ├── search.py      LLM call, page extraction, demo cache
│   │   ├── classify.py    deterministic keyword → axis (no LLM)
│   │   ├── data.py        people.json + results.json loaders
│   │   ├── llm.py         Anthropic client, key round-robin
│   │   └── models.py      pydantic schemas
│   ├── data/
│   │   ├── manifestos/    labour.md, conservative.md, libdem.md, snp.md, plaid.md
│   │   ├── axes.json      6 axes × 5 parties — verbatim quotes + compass positions
│   │   ├── results.json   timeline entries by topic and by person
│   │   └── people.json    3 demo MP profiles
│   └── demo_cache/        15 pre-warmed responses for offline demo
└── frontend/
    ├── src/
    │   ├── App.jsx         root layout + search state
    │   ├── Map.jsx         Leaflet 4-nation interactive map
    │   ├── PartyCard.jsx   topic card + person empty card
    │   ├── DetailPanel.jsx right panel — timeline + person profile
    │   ├── api.js          backend fetch + mock toggle
    │   └── styles.css      glass aesthetic
    └── public/
        └── uk-nations.geojson   ONS 4-nation boundaries

---

Data sources

Source	Licence	Used for
2024 UK party manifestos (Labour, Conservative, Lib Dem, SNP, Plaid Cymru)	Public political content	Manifesto text, verbatim citations
ONS Open Geography Portal	Open Government Licence	UK nations GeoJSON
Democracy Club	CC BY 4.0	MP profiles, candidate data
UK Parliament	Open Parliament Licence	Constituency election results

---

Team

Builder	Role
Saliha (@saliha006)	Backend — FastAPI, LLM integration, demo cache
Abdulla (@Abdulla-AlBassam)	Database & Frontend — JavaScript (React), Leaflet, UI, prompt engineering
Elya (@ElyaRaza)	LLM prompts — citation quality, extraction, prompt caching
Maks (@maksymkhomitskyi)	Data — manifestos, axes, MP profiles, demo script
Aws (@awszaman)	Joker — paired across all lanes