Odoo RAG Chatbot System

Project Overview

Many organizations store critical knowledge inside Odoo (documents, attachments, internal files), but accessing that information usually requires manual searching, navigation, or support requests. This project aimed to unlock that knowledge through natural language, allowing users to simply ask questions and receive accurate, context-aware answers.

I designed and delivered an AI-powered Retrieval-Augmented Generation (RAG) chatbot that enables users to chat directly with Odoo-stored documents, while ensuring data accuracy, scalability, and maintainability.

Problem Statement

• Knowledge was fragmented across Odoo documents and difficult to retrieve quickly.

• Users relied on manual searches or support teams to find information.

• Traditional chatbots failed due to hallucinations or lack of document grounding.

• Document updates were frequent, requiring reliable synchronization with the AI system.

Solution

I built a RAG-based conversational system tightly integrated with Odoo, designed around automation, accuracy, and real-world usage.

The solution:

• Ingests and indexes Odoo documents automatically

• Retrieves only relevant, context-specific content

• Generates grounded responses based strictly on retrieved data

• Stays synchronized with document updates and deletions

The result is a chatbot that behaves less like a generic assistant and more like a knowledge interface for Odoo.

System Architecture & Approach

The system was designed as a modular, event-driven pipeline:

1. Document ingestion from Odoo (attachments, files, updates)

2. Automated processing workflows triggered via webhooks

3. Vector embedding and storage in a dedicated vector database

4. Contextual retrieval based on user queries

5. LLM-based response generation grounded in retrieved content

6. Cleanup and re-sync logic to handle document changes reliably

Special care was taken to:

• Avoid stale embeddings

• Prevent irrelevant context injection

• Balance latency, accuracy, and cost

Key Responsibilities

• Designed the end-to-end RAG architecture for Odoo document querying

• Built automated ingestion and synchronization workflows using n8n

• Defined the vector database schema and embedding strategy

• Implemented contextual retrieval logic to improve answer relevance

• Designed conversational UX logic to guide users toward clear, answerable queries

• Integrated webhook-based triggers for document updates and cleanup

• Conducted performance benchmarking and retrieval quality testing

• Tuned chunking, embeddings, and prompts to reduce hallucinations

Technologies Used

• Odoo — document source and business system

• n8n — workflow automation and orchestration

• Supabase Vector — vector storage and similarity search

• OpenAI / Gemini — embeddings and response generation

• Python — processing, validation, and testing logic

• Webhooks — real-time synchronization triggers

Challenges & Learnings

• Ensuring data freshness was more critical than model choice

• Poor chunking quickly degraded answer quality

• RAG systems require strong retrieval discipline to avoid hallucinations

• Automation and cleanup workflows are essential for production readiness

This project reinforced that RAG is a systems problem, not just an LLM problem.

Outcome & Impact

• Enabled instant access to Odoo knowledge through natural language

• Reduced reliance on manual searches and support requests

• Delivered a scalable foundation that can be extended to new document types

• Provided a production-ready RAG pattern reusable across other systems