🗳️ Election RAG — Open-Source Retrieval-Augmented QA with Qdrant

An end-to-end Retrieval-Augmented Generation (RAG) system that allows users to ask questions about political party election plans and receive grounded, source-backed answers.

This project uses:

• Qdrant Cloud as the vector database
• Local embeddings (HuggingFace / Sentence Transformers)
• Local LLM via Ollama (no paid APIs)
• FastAPI backend
• Zero paid APIs

Designed as a realistic, minimal, and well-structured RAG system, suitable for educational use, civic transparency, and as a reference architecture.

---

✨ Features

• 📄 Ingests multiple PDF documents (party programs, manifestos, plans)
• 🔍 Semantic search using vector embeddings
• 🧠 Grounded answers using RAG (no fine-tuning)
• 🔗 Source citations for every answer
• ☁️ Uses Qdrant Cloud (free tier supported)
• 🧠 Fully local inference (zero OpenAI / paid LLM APIs)
• 🚀 FastAPI service
• 🧱 Clean, modular Python architecture

---

🧠 Architecture Overview

PDFs
↓
Text Extraction
↓
Chunking (512 tokens)
↓
Local Embeddings (Sentence Transformers)
↓
Qdrant Cloud (Vector Storage)
↓
Retriever (Top-K similarity search)
↓
Local LLM (Ollama)
↓
FastAPI (/query endpoint)

---

🧰 Tech Stack

Component	Technology
Vector DB	Qdrant Cloud
Embeddings	HuggingFace (bge-small-en-v1.5)
LLM	Ollama (llama3)
RAG Framework	LlamaIndex
API	FastAPI
PDF Parsing	PyPDF
Language	Python 3.10+

---

📁 Project Structure

election-rag/
├── api/                # FastAPI app
├── ingest/             # PDF ingestion & indexing
├── rag/                # Query & LLM logic
├── config/             # Environment-based settings
├── data/
│   └── raw/pdfs/       # Input PDFs
├── scripts/            # Dev & testing scripts
└── README.md

---

🚀 Getting Started

1️⃣ Prerequisites

• Python 3.10+
• Ollama (for local LLM)
• A Qdrant Cloud account (free tier)

---

2️⃣ Install Ollama (local LLM)

brew install ollama
ollama serve
ollama pull llama3

Verify:

ollama run llama3 "Hello"

3️⃣ Clone the repository

git clone https://github.com/your-username/election-rag.git
cd election-rag

4️⃣ Create virtual environment

python3 -m venv .venv
source .venv/bin/activate

5️⃣ Install dependencies

pip install --upgrade pip
pip install -r requirements.txt

Or manually:

pip install \
  fastapi uvicorn \
  llama-index \
  llama-index-llms-ollama \
  llama-index-embeddings-huggingface \
  llama-index-vector-stores-qdrant \
  qdrant-client \
  sentence-transformers \
  pypdf \
  python-dotenv

☁️ Qdrant Cloud Setup

1. Create a free Qdrant Cloud cluster
2. Copy:
   • Cluster URL
   • API key

Create a .env file:

QDRANT_URL=https://your-cluster.qdrant.io
QDRANT_API_KEY=your_api_key_here

📄 Add Your PDFs

Place election plans here:

data/raw/pdfs/

🧱 Ingest Documents (One-Time Step)

This: • Loads PDFs • Chunks text • Generates embeddings • Stores vectors in Qdrant Cloud

python -m ingest.run_ingest

🧪 Local Query Test (Optional)

python -m scripts.dev_query

🌐 Run the API

uvicorn api.app:app --reload

Server runs at: http://127.0.0.1:8000

📘 API Usage

Ask a Question

POST /query
Content-Type: application/json

Body

{
  "question": "What does Party A propose about healthcare?"
}

Sample Response

{
  "answer": "Party A proposes increasing healthcare funding...",
  "sources": "party_a_2025.pdf (page 12)"
}

⚠️ Notes on Accuracy & Safety

• Answers are grounded in retrieved documents
• Local LLMs may hallucinate — always verify sources
• Intended for educational and informational use