Robert-Owoo/Ai_10022300108

A.i Chatbot for Ghana's election and budget # Academic City RAG Chatbot (Manual RAG Implementation) **Student Name:** Robert George Owoo **Index Number:** 10022300108 ## Overview This project implements a **Retrieval-Augmented Generation (RAG)** chat assistant for Academic City using: - **Ghana Election Results CSV** (provided dataset) - **MOFEP 2025 Budget Statement PDF** (provided dataset) **Important constraint satisfied:** No end-to-end RAG frameworks (no LangChain, LlamaIndex, or pre-built RAG pipelines). Core RAG components are implemented manually: - Data cleaning - Chunking (with configurable size/overlap) - Embedding pipeline (Sentence Transformers) - Vector similarity + top-k retrieval - **Hybrid retrieval** (keyword BM25 + vector) - Prompt construction + context window management - Full pipeline logging + UI display of retrieved chunks/scores/final prompt ## Architecture ```mermaid flowchart LR A[User Query] --> B[Query Prep] B --> C[Hybrid Retrieval (BM25 + Vector)] C --> D[Top-k Chunks + Scores] D --> E[Context Selection (truncate / filter)] E --> F[Prompt Template (hallucination controls)] F --> G[LLM] G --> H[Answer + Citations] C --> L[Stage Logs] E --> L F --> L G --> L ``` ## What’s inside - **Streamlit UI**: `app.py` - **Index building**: `scripts/build_index.py` - **Chunking experiments**: `scripts/run_chunking_experiments.py` - **RAG vs pure-LLM evaluation**: `scripts/evaluate_rag_vs_llm.py` - **Manual experiment log template**: `experiments/manual_log.md` - **Architecture explanation**: `docs/architecture.md` ## Setup ### 1) Create and activate a virtual environment ```bash python -m venv .venv .\.venv\Scripts\activate ``` ### 2) Install dependencies ```bash pip install -r requirements.txt ``` ### 3) Configure API key (for generation) This app uses an OpenAI-compatible API for generation. - Windows PowerShell: ```powershell $env:OPENAI_API_KEY="YOUR_KEY" $env:OPENAI_MODEL="gpt-4o-mini" ``` If you don’t have an API key, you can still run indexing + retrieval, but generat …