IP Automation AI, SaaS & digital platforms
IP360
AI-driven contract automation with RAG and knowledge graphs
For 7 Miles Global, Valens built IP360's AI application for document ingestion, data extraction and contract preparation, and integrated it with an existing production system of record. RAG, knowledge graphs and human review work together, so contracts are prepared faster without giving up control.
The problem
Contract and talent data arrived in unstructured documents and had to be entered by hand into a legacy system whose business rules were complex and only partly documented. Processing was slow and error-prone.
Our delivery
A new AI application that ingests documents, extracts contract fields with a hosted LLaMA model and prepares contracts for human review. It connects through backend-only APIs to a stabilised legacy application that remains the system of record.
How we worked
Project context
7 Miles Global builds data- and automation-driven platforms for complex business operations, and IP360 is part of its effort to modernise contract and talent management. The job was to introduce a new AI application while stabilising a legacy production system. The two applications run on separate domains, technology stacks and Neo4j databases, yet had to work as one system for users.
Delivery scope
In the new application: document ingestion and extraction, n8n-orchestrated processing workflows, AI-assisted contract preparation with user and admin review, an AI chat for data gathering and a document-aware RAG chat. In the legacy application: bug fixing and stabilisation, correction of dependency-driven business rules, support for users with multiple profiles, Microsoft Entra ID single sign-on, backend API changes for secure integration, and AI extraction of completion reports. Frontend, backend and QA across both.
How we built it
Documents enter through LightRAG, and a hosted LLaMA model extracts the required contract fields. Neo4j models the relationships between entities, and Qdrant stores vector embeddings for semantic search. Users review and edit each draft, and an admin steps in when required entities or values are missing. Approved contracts are created through the legacy API and cached in Redis, so they are available in both applications. In the legacy app, Azure AI Intelligence reads completion reports but has no access to the database: backend APIs apply the business rules and create the entities. The two applications talk only through backend services and share one sign-on, with a single active profile across both.
Tech stack
- Next.js
- NestJS
- React
- .NET
- Neo4j
- Qdrant
- Redis
- LightRAG
- LLaMA
- n8n
- Azure AI Intelligence
- Microsoft Entra ID
Impact
Manual document processing and data entry dropped significantly, data stayed consistent across both systems, and the legacy application became more stable. AI now handles extraction and drafting, while people keep the final decision on every contract.
- Faster
- contract processing
- Visibility
- Centralized visibility into assets and rights
- One system
- two applications, one sign-on
Have a build that looks like IP360?
We pair product, engineering, and AI delivery to turn shaped problems into working products that stay reliable in operation.
Next project
Hanover Research
Research Platform
