← Back to Portfolio Chatbot · Finance

O2C SOP Chatbot

A ChatGPT-style assistant trained on a 150-page Order-to-Cash SOP

Overview

Authenticated users ask questions about the Order-to-Cash (O2C) process in NetSuite, and the bot retrieves the relevant sections — and screenshots — directly from the source SOP document, answering with step-by-step instructions and page citations rather than a generic response.

How it works

FastAPI backend with a React frontend, using Groq's Llama 3.3 70B for generation. Retrieval is hybrid: MiniLM sentence embeddings combined with BM25 keyword search, so answers stay grounded in the actual 150-page SOP rather than the model's general knowledge. If the embedding model can't load, the app automatically falls back to BM25-only retrieval instead of failing.

Deployed as a Docker Space on Hugging Face.

Access & roles

Visit Project