Ghana Investment Insights Portal

Abstract

This project presents a business-focused portal for exploring Ghana's investment landscape using official Bank of Ghana reports. Users can upload the latest PDF report and ask questions about sector trends, macroeconomic indicators, and regulatory changes. The system uses a Retrieval-Augmented Generation (RAG) pipeline with ChromaDB, HuggingFace embeddings, Groq LLM, and a custom Streamlit interface for a seamless, data-driven experience.

Introduction

Investors and policymakers need timely, reliable insights to make informed decisions. This portal bridges the gap between raw financial data and actionable intelligence by allowing users to interact with official Bank of Ghana reports through natural language queries. The solution is designed for clarity, transparency, and business usability.

Methodology

Tools, Frameworks, and Libraries

• Python 3.10+: The primary programming language for the project.
• Streamlit: Used to build a clean, interactive, and business-focused web interface for user interaction.
• PyPDF2: For extracting text from PDF reports.
• LangChain: Provides the pipeline for text chunking, embedding, and retrieval.
• ChromaDB: A persistent vector database for storing and searching document embeddings efficiently.
• HuggingFace Transformers & Sentence Transformers: Used for generating dense vector embeddings of text chunks. Specifically, the sentence-transformers/all-MiniLM-L6-v2 model was chosen for its balance of speed, accuracy, and resource efficiency in semantic search tasks.
• Groq LLM (Llama 3.1-8b-instant): The large language model used for generating answers. Chosen for its strong performance, cost-effectiveness, and ability to follow custom system prompts.
• python-dotenv: For secure management of API keys and environment variables.

Implementation Process

• PDF Ingestion: A Bank of Ghana (Ghana's central bank) PDF report is uploaded into the data/ folder. The system uses PyPDF2 to extract and concatenate text from all pages.
• Text Chunking: The extracted text is split into overlapping chunks using LangChain's RecursiveCharacterTextSplitter (chunk size: 1000, overlap: 200) to preserve context and improve retrieval accuracy.
• Embedding & Vector Storage: Each chunk is embedded using HuggingFace's MiniLM model (all-MiniLM-L6-v2) and stored in ChromaDB. This enables fast, semantic similarity search for user queries.
• Retrieval & Prompting: When a user submits a question, it is embedded and compared to all stored chunks. The top-k most relevant chunks are retrieved. These are combined with a business-focused system prompt and sent to Groq LLM (Llama 3.1-8b-instant) for answer generation. The prompt ensures the LLM only uses information from the uploaded report and responds in a clear, professional style.
• User Interface: The Streamlit UI is styled as a business portal, with branding, color, and layout tailored for professional users. Users can type questions, view answers, and see a history of their queries and responses.

---

Screenshots

---

Demo

Live Link: https://ghana-investment-insights.streamlit.app/

---

Features

• Upload and analyze official Bank of Ghana PDF reports
• Ask questions about Ghana's economy, sectors, or regulations
• Get answers grounded in real, up-to-date data
• Clean, business-focused web interface (Streamlit)
• Persistent vector search with ChromaDB

---

Example Questions

• What sectors are growing fastest in Ghana?
• What is the current inflation rate?
• What are the latest trends in the banking sector?

---

Project Structure

RAGchatbot/
├── app.py                # Streamlit UI (Ghana Investment Insights Portal)
├── requirements.txt      # Python dependencies
├── .env                  # For your GROQ_API_KEY
├── data/                 # Bank of Ghana PDF report is placed here
├── screenshots/          # App screenshots
├── src/
│   ├── main.py           # Core pipeline: PDF ingestion, embedding, retrieval

---

Installation & Usage

1. Clone the repository
2. Install dependencies:

   pip install -r requirements.txt

3. Set up your environment:
   • Copy .env.example to .env and add your Groq API key:

     GROQ_API_KEY=your-groq-api-key-here

   • Place your Bank of Ghana PDF in the data/ folder (e.g., Summary-of-Economic-and-Financial-Data-May-2025.pdf).
4. Run the portal:

   streamlit run app.py

---

License

MIT