Axonaut

This case study demonstrates how we transformed Axonaut, a leading French business management software serving 189,000+ SMEs, into an AI-powered platform that sets new standards for customer experience and operational efficiency.

 

Project Objectives

 

1. Autonomous AI Copilot Agent — The Centerpiece of Digital Experience

Transform the traditional support chatbot into a powerful autonomous AI agent that acts as a personal assistant for every Axonaut user. Customers can perform any action in their account directly through natural conversation — no menus or manual entry required. The agent analyzes screenshots uploaded by users (e.g., receipt photos), extracts the data, and automatically creates corresponding records. It also answers questions about the platform's features and accounting procedures by consulting the knowledge base.

Key capabilities include:

  • Natural-language business actions: Execute any operation in the Axonaut dashboard (create invoices, add expenses, update records) through conversational commands

  • Vision-powered automation: Upload a screenshot or photo, and the agent analyzes it to understand context and perform appropriate actions

  • Knowledge-grounded answers: Get accurate, contextual responses drawing from the accountant knowledge base for questions about platform features, settings, and accounting procedures

  • Enterprise observability: Axonaut administrators can monitor all agent activities through AWS CloudWatch — tracking which user invoked which tool, when, and with what results — ensuring full auditability and compliance

  • Full conversation history: Sessions maintain complete context across interactions, while long-term history is preserved for audit and reference

 

2. Intelligent Document Processing — From Manual Entry to Automated Extraction

Eliminate manual data entry by implementing an AI-powered pipeline that extracts structured financial data from invoices and receipts, producing accounting-ready records with French PCG code suggestions.

 

3. AI-Powered Business Communication

Enable users to generate professional business emails with tone adaptation, translation, and summarization capabilities — all within the Axonaut ecosystem.

 

4. Automated Privacy Compliance

Deploy intelligent PII detection and redaction to protect sensitive customer data in uploaded documents.

 

Challenges

 

1. Building a Production-Grade Autonomous Agent

Creating an agent capable of executing real business actions required solving complex challenges: multi-step reasoning with controlled tool execution, vision understanding for screenshot analysis, session-aware conversation flow with memory, and granular permission management per user. The solution demanded tight integration between Bedrock AgentCore, LangGraph orchestration, and Axonaut's business logic through MCP — while maintaining security and user-specific authorization at every step.

 

2. Performance Optimization for Document Processing

Initial Textract-based PDF processing showed unacceptable latency for production use. We rearchitected the pipeline using native Python PDF parsing combined with Claude Sonnet for document analysis, implementing Lambda warm-up strategies to eliminate cold start delays and achieve response times suitable for real-time user interactions.

 

3. Multi-Knowledge Domain RAG Architecture

Implementing retrieval-augmented generation across three separate knowledge bases (client, visitor, accountant) with proper access permissions and session continuity required careful orchestration. Since Terraform did not yet support the latest Bedrock Knowledge Base resources, we deployed components using custom Python scripts, maintaining infrastructure as code principles throughout.

 

4. Security and Tenant Isolation

With multiple user types accessing different capabilities, implementing proper tenant isolation, granular permission controls, and comprehensive audit trails required fine-grained IAM policies and API-level security mechanisms.

 

Technologies

 

1. AWS AI & ML Services

Our solution leverages the full depth of AWS AI services:

  • Amazon Bedrock — Foundation platform for generative AI, model inference, and knowledge base retrieval

  • Amazon Bedrock AgentCore — Agent orchestration framework for multi-step reasoning, tool execution, and conversational memory management

  • Bedrock Knowledge Bases — RAG implementation across three knowledge domains with vector search

  • Bedrock AgentCore Memory — Short-term conversational memory for session continuity

  • Claude Sonnet family — Foundation models for text generation, vision analysis, and document understanding

  • Amazon Textract — OCR capabilities for document extraction in hybrid processing paths

  • Amazon Comprehend — PII entity detection and language identification

  • Amazon Translate — Multilingual translation workflows

 

2. AWS Cloud Infrastructure

Production-grade serverless architecture ensures scalability and cost efficiency:

  • Amazon API Gateway — Secure endpoints with usage plans and throttling

  • AWS Lambda — Business logic orchestration with provisioned concurrency

  • Amazon OpenSearch Serverless — Vector storage for knowledge base embeddings

  • Amazon S3 — Document storage and conversation archives

  • AWS IAM — Fine-grained access control and service roles

  • Amazon CloudWatch Logs — Observability and diagnostics

 

3. AI Orchestration & Governance

  • LangGraph — Stateful multi-step orchestration with human-in-the-loop confirmation workflows

  • MCP (Model Context Protocol) — Secure integration layer enabling the agent to execute business tools both on the Axonaut platform and through our custom AWS services

  • Session-aware conversation management with split short-term/long-term memory

  • Comprehensive audit logging tracking every tool invocation, user action, and system response

 

Process

The engagement began when Axonaut approached Noveo with a requirement to implement four distinct AI services: an intelligent support copilot with knowledge base integration, an AI email assistant, document processing automation, and a PII redaction service. Our team delivered all four services within the agreed timeline, establishing production-grade APIs with proper security, observability, and scalability.

Following successful completion of the initial scope, Axonaut recognized the opportunity to transform their support copilot into something far more powerful. We proposed extending the chatbot into an autonomous AI agent capable of executing real business actions — integrating the document processing capabilities directly into the conversational experience. The client embraced this vision, and we proceeded to build the advanced AI Copilot Agent that now serves as the centerpiece of Axonaut's digital experience.

The project was delivered by a cross-functional Noveo team comprising Solutions Architects, AI/ML Engineers, Backend Engineers, and DevOps specialists.

 

Results

  • Deployed a production-ready Autonomous AI Copilot Agent that enables Axonaut users to execute any dashboard action through natural conversation — from creating invoices to managing expenses — with vision-powered screenshot analysis that automatically extracts and records data from user photos

  • Achieved 4x cost reduction on document processing by replacing a third-party service (which charged premium subscription fees) with our optimized in-house solution built on AWS Textract and Bedrock

  • Established enterprise-grade observability with comprehensive audit logging: every user action, tool invocation, and system response is tracked with full traceability — enabling Axonaut to monitor usage patterns, ensure compliance, and continuously improve the experience

  • Delivered four production AI services with production-grade API posture, security controls, and comprehensive monitoring

  • Implemented multi-knowledge domain RAG with session continuity across three distinct user types (clients, visitors, accountants)

  • Built hybrid OCR + GenAI document processing that transforms invoices into accounting-ready data with French PCG code suggestions

  • Delivered automated PII redaction supporting mask, remove, and highlight modes for compliance workflows

  • Achieved serverless scalability with cost-efficient usage-based pricing across all services

 

Get your estimation