Building a Retrieval-Augmented Generation (RAG) Chatbot

🔎 What is a RAG Chatbot?

RAG stands for Retrieval-Augmented Generation. It combines:

• 🔹 Retriever – finds the most relevant pieces of data for a query
• 🔹 Generator – uses an LLM to create a natural-language answer

Instead of relying only on pre-trained knowledge, the bot dynamically looks into your dataset first, then crafts an answer based on that context.

⚙️ Tech Stack

• Python for scripting and data processing
• Pandas for reading & cleaning CSV data
• Sentence Transformers (all-MiniLM-L6-v2) for embeddings
• FAISS for similarity search (retrieving relevant text chunks)
• Flask for the minimal web UI
• OpenRouter API to query a language model for response generation

🛠️ Implementation Steps

1. Read BotData.csv and parse the ChunkContentDto.Content field
2. Pre-process numeric/date fields for analytics queries
3. Create embeddings with SentenceTransformer and store them in a FAISS index
4. Detect if the user’s question is numeric (sum, average, etc.) or text-based
5. For numeric queries → compute directly and optionally format using the LLM
6. For text queries → retrieve relevant chunks & ask the LLM for a concise answer
7. Serve a very light Flask HTML interface for asking questions

📸 UI Previews

Chatbot Main Page

Query Input & Retrieval

Answer Display & Context

📜 Key Code Snippet

# Build FAISS index
data_embeddings = model.encode(blocks, convert_to_numpy=True)
dimension = data_embeddings.shape[1]
index = faiss.IndexFlatL2(dimension)
index.add(data_embeddings)

# Retrieve context
def get_context(query, top_k=20, max_chars=12000):
    query_embedding = model.encode([query], convert_to_numpy=True)
    distances, indices = index.search(query_embedding, top_k)
    return "\n\n".join([blocks[i] for i in indices[0]])[:max_chars]

# Flask route
@app.route("/", methods=["GET", "POST"])
def home():
    ...
      

Full code available on request or in the project files.

💡 Lessons Learned

• Managing embeddings efficiently is crucial for performance
• Prompt design matters — concise context improves answer quality
• Logging errors from API calls saves debugging time
• Even a minimal Flask UI makes testing far easier than CLI