OvexSync ERP

Client

The client is a FinTech company that needed to replace fragmented internal tools with one system for core business workflows. The project covered HR, recruitment, sales, lead generation, analytics, and operations, with a particular focus on reducing manual work in recruitment and keeping data connected across teams.

The product therefore had two simultaneous requirements: function as a modular enterprise ERP across six business domains and provide a dedicated AI recruitment suite for matching, scoring, search, competency analysis, and pre-screening.

Key Highlights

Challenge

The starting point was a fragmented internal landscape: core business workflows lived in different tools, while recruitment still required significant manual review and matching. The engineering challenge was to bring those domains together without turning the product into one tightly coupled application.

• Define domain boundaries. Separate the six business domains while still allowing them to exchange data.
• Go beyond keyword filtering. Build an AI recruitment suite that remains flexible across model providers.
• Support durable workflows. Handle long-running processes that may span multiple services and must recover from intermediate failures.
• Integrate broadly and securely. Connect external tools and data sources while maintaining role-based access across the platform.
• Design responsibly. Architect for enterprise workloads without making unverified performance claims.

Solution

We implemented OvexSync as a modular microservices-based platform and assigned each major technology a distinct role in the architecture.

• Microservices architecture. Separate business domains can be deployed and scaled independently where needed, while remaining connected through defined interfaces and messaging.
• Python AI/ML layer. Matching, scoring, and competency-analysis logic lives in the ecosystem best suited to model-driven features.
• LiteLLM. One interface for working with multiple language-model providers; it handles model access rather than search.
• Meilisearch. Semantic candidate discovery and retrieval based on relevance rather than literal keyword matching.
• Temporal. Durable orchestration for multi-step recruitment and operational workflows.
• RabbitMQ, Redis, PostgreSQL, Docker, and AWS. Messaging, caching, relational persistence, containerized delivery, and cloud infrastructure for the connected platform.
• Next.js + Fastify/Node.js. The user-facing web application and API services, with role-based access and external integrations.

Process

The project was organized around domain boundaries and the recruitment workflow rather than a generic feature-by-feature build sequence.

1. Map the domains. Mapped the six business domains and defined how data and responsibilities should be separated between services.
2. Build the backend foundation. Built the shared backend foundation, messaging, persistence, caching, and containerized delivery setup.
3. Implement the AI suite. Implemented the AI recruitment suite in Python, with LiteLLM for model access and Meilisearch for semantic search.
4. Add workflow orchestration. Added Temporal orchestration for multi-step workflows that need durable execution and recovery.
5. Build the interface and integrations. Developed the Next.js interface, role-based access, and the documented external integrations.
6. Deploy and support. Ran QA, deployed through CI/CD on AWS, and continued supporting the platform as it evolved.

Result

• Unified business domains. Six core business domains operate within one platform instead of being represented as separate products in the solution architecture.
• Complete recruitment suite. The AI recruitment suite combines matching, semantic search, scoring, competency analytics, and automated pre-screening in the same recruitment workflow.
• Durable process handling. Long-running processes are handled through durable workflow orchestration rather than ad-hoc chained requests.
• Provider-independent AI. The AI layer is not locked to a single model provider, and semantic search is separated from model routing as a dedicated capability.
• Enterprise-ready access model. The platform has an integration layer and role-based access model that support its use across multiple internal teams.