TL;DR

Seldon is a comprehensive web platform designed for a structural engineering consultancy to modernize their sales and administrative workflows. What was once a fragmented, manual process—sorting through emails, documents, and outdated ERP interfaces—became a smooth, semi-automated system. At the heart of Seldon is an AI agent powered by a RAG + MCP architecture, trained on incoming documents and emails, enabling the team to quickly understand complex requests and act with confidence.

My Role

I acted as Product Owner, Architect, and Full-Stack AI Developer, guiding the project from concept to execution. I worked closely with stakeholders to define the roadmap and translate their needs into actionable technical solutions. I designed the architecture, implemented backend microservices and the frontend interface, and built the AI stack—including RAG pipelines, LangChain agents, and the MCP server. I also tackled legacy challenges, reverse-engineering ERP APIs to ensure smooth system integration. The project required balancing cutting-edge AI technology with practical usability for a traditional industry team.

Tech Stack

The project was built on a modern web stack that allowed us to move fast without sacrificing scalability or performance.

The frontend uses Next.js, providing a dynamic and responsive interface. The backend is powered by NestJS microservices, orchestrated with Redis and GraphQL for high-performance data handling. We leveraged ArangoDB and OpenSearch for flexible data storage and retrieval, while the MCP server orchestrates AI-driven operations.

On the AI side, LangChain and OpenAI APIs handle natural language understanding, and a custom RAG pipeline ensures contextual comprehension of documents.

Integrations include Microsoft Graph API, SharePoint Online, and the legacy ERP system, which we interfaced with despite its undocumented APIs.

The stack allowed us to seamlessly combine modern web technologies with AI capabilities to deliver a robust, full-stack solution.

Challenges & Breakthroughs

Working with a non-technical client required patience, creativity, and translation of complex concepts into actionable solutions. Together, we shaped a product vision that was both ambitious and achievable. Developing a domain-specific RAG agent capable of understanding civil engineering projects, contracts, and legal documents required iterative fine-tuning and careful orchestration of AI pipelines. On the frontend, we delivered a complete UI/UX overhaul, replacing a cumbersome legacy interface with an intuitive, user-friendly experience that empowered the team to adopt modern tools confidently.

Key Features Built

AI Document Intelligence Agent

The customer handled a huge volume of emails and attachments, often containing complex civil engineering requests and legal documents. I built an AI agent powered by a RAG pipeline that reads and understands these documents, extracts key entities, and answers internal questions. The agent can even trigger internal workflows via the MCP server, helping the team quickly respond to client requests with confidence and accuracy.

Legacy System Integrations

The company’s legacy ERP and SharePoint systems were critical but poorly documented, making data access slow and fragmented. I reverse-engineered the ERP APIs and implemented seamless synchronization with SharePoint and Office365, ensuring that documents and communications flow smoothly across platforms. This integration preserved the team’s existing processes while unlocking opportunities for automation and better collaboration.

Modern Workflow & User Interface

The old interface was cluttered and unintuitive, slowing down the team’s daily operations. I completely redesigned the UI/UX, creating a clean, responsive interface that simplifies navigation and highlights the most important tasks. Combined with AI-driven insights, this modern workflow allows the team to operate efficiently, quickly access relevant information, and adopt technology without friction.

Outcome

The team can handle complex requests with unprecedented speed and confidence, relying on AI-driven insights while maintaining control over workflows. The platform bridges the gap between legacy systems and modern interfaces, empowering a traditionally low-tech team to embrace advanced technology without disruption. It’s incredibly rewarding to see a small team in a conservative industry adopt a modern, AI-enhanced workflow that genuinely changes how they work.