Local Document Q&A Chatbot with Mock Embeddings

This project implements a simple Retrieval-Augmented Generation (RAG) pipeline using custom mock embeddings instead of pre-trained ones. It enables users to load a local .txt document and ask questions, retrieving relevant chunks and generating simulated LLM responses.

🧠 Project Overview

A command-line chatbot that:

• Loads a local text file
• Splits the content into chunks
• Generates mock embeddings
• Stores chunks in a simple in-memory vector store
• Retrieves the top K relevant chunks based on cosine similarity
• Augments the query with retrieved content
• Simulates an LLM response using a mock function

This project is ideal for learning the inner workings of document retrieval, vector similarity, and chatbot logic — without relying on external APIs or heavy libraries.

---

🚀 Features

• 📂 Load any .txt file from your local machine
• ✂️ Custom chunking function to split large text
• 🧮 Mock embedding generator using ASCII transformations
• 🧠 Simple vector store with cosine similarity retrieval
• 🤖 Augmented prompt construction and fake LLM response generation
• 🔁 Continuous Q&A chat loop until user exits

---

🛠️ Tech Stack

• Python (Standard Library)
• No external libraries or frameworks

---

🧪 How It Works

1. Load & Chunk Document
   • Text is split into readable-sized chunks (default 100 chars).
2. Generate Mock Embeddings
   • Each character in the chunk is converted to a float using a simple formula.
3. Vector Storage
   • Chunks and embeddings are stored in a basic list of tuples.
4. Query & Retrieval
   • User's question is also embedded and cosine similarity is computed.
   • Top K relevant chunks are retrieved.
5. Prompt Augmentation
   • A fake LLM prompt is created using retrieved context and query.
6. Mock Response
   • Simulated answer is printed with reference to matched content.