Overview

Meridpath is a logistics enterprise that runs its core operations on a legacy SOAP-based ERP that exposed inventory, order, and shipment data only through aging XML web services. The company wanted an AI agent in front of that system so staff could query records and trigger updates in plain language. The agent could not call the SOAP services directly, and the ERP sent back inconsistent, loosely typed responses that no model could trust. The ERP also enforced session-based authentication and handled only a handful of concurrent calls, so direct agent traffic risked timeouts and corrupt writes. Bacancy built an integration layer that sits between the agent and the ERP. It converts each agent request into a SOAP transaction, validates it, runs strict schema checks before any write executes, queues and retries calls to respect the ERP’s limits, and returns clean, structured, auditable data the agent acts on with confidence.

Technologies Used

python
LangChain
FastAPI
RabbitMQ
Docker
PostgreSQL

Project Highlights

checkmark

WSDL-Driven SOAP and XML Integration Layer

checkmark

AI Agent Tool and Action Interface

checkmark

Validated and Auditable ERP Transaction Gateway

checkmark

Asynchronous Message Queue with Retry Buffering

The Challenges

1

The AI agent could not produce well-formed SOAP and XML payloads the ERP required.

2

Non-deterministic agent outputs risked corrupting transactions inside the fragile legacy ERP system.

3

The legacy ERP had limited concurrency, slow responses, and fragile session-based authentication.

4

Inconsistent ERP data structures resisted mapping into clean schemas the agent could consume.

Solutions by Bacancy

1

With our AI integration services, we addressed the challenge of connecting modern AI agents with a SOAP-based ERP through a WSDL-driven translation layer. The solution automatically translated JSON requests into SOAP messages and converted XML responses into structured, agent-ready data for seamless communication.

2

Our experts added a validation gateway that schema-checks each agent request, enforces field types and value ranges, and validates required inputs before processing. This ensures only accurate, well-formed payloads reach the ERP system and prevents invalid writes from impacting business data.

3

We wrapped the ERP behind a RabbitMQ queue with retry and backoff logic and a managed session-token pool that refreshes credentials automatically. This absorbs the system’s low concurrency, slow response times, and fragile authentication instead of pushing those limits onto the agent.

4

Our team profiled the ERP’s inconsistent fields and built normalized schema mappings that standardize naming, types, and structure. The agent now receives predictable, well-typed objects on every call rather than parsing messy, irregular XML each time it reads data.

Core Features

checkmark

Schema Validation Engine

checkmark

Session Token Manager

checkmark

Transaction Audit Logging

checkmark

Automatic Retry Handling

No. of Resource

04

No. of Resource

Time Frame

September 2025 - January 2026

Time Frame

Project Snapshot

Outcomes

100% of ERP writes validated before they execute

0 corrupt transactions reaching the legacy ERP

3 ERP workflows now triggered in plain language

0 manual data entry between agent and ERP

100% of ERP transactions captured for audit

24/7 agent access to live ERP data

Technical Stack

Backend PythonPydanticZeepLangChainFastAPI
Database PostgreSQL
In-memory Data Store Redis
Cloud Infrastructure AWS (ECS, CloudWatch)
Architecture Event-Driven MicroservicesContainerized
API Communication REST APIsSOAP/XMLAMQP
Project & Issue Tracking Jira

Experience With Bacancy

2500+ Projects Experienced Innovation with Bacancy!

Get access to an experienced team of developers and engineers from Bacancy, handpicked to ace your goals. Kickstart within 48 hours, no-risk trial.

Book a 30 min call

14+

Years of Business Experience

1458+

Happy Customers

12+

Countries with Happy Customers

1050+

Agile Enabled Employees

How Can We Help?