Secure web access
Classification of 38,000 documents
← All projects
DELIVERY PRACTICE · Artificial intelligence
Enterprise AI in a controlled environment
Local language models, RAG over corporate documents, a secure web interface, access controls and answer-quality safeguards without sending data to external services.
Year2026Timeline16 weeksEngineering company · 38,000 documents
Initial situation
Virtek’s in-house AI platform
Engineers spent up to half an hour finding requirements in project documentation, while policy prohibited sending data to public AI services.
Architecture
Enterprise AI in a controlled environment
RAG orchestration
Local DeepSeek and Qwen models with request routing
Vector index
RAG search with mandatory source citations
Local LLMs
Role-based access, logs, quality evaluation and GPU monitoring
GPU layer and audit
Role-based access, logs, quality evaluation and GPU monitoring
Solution scope
Engineering company · 38,000 documents
- Classification of 38,000 documents
- Local DeepSeek and Qwen models with request routing
- RAG search with mandatory source citations
- Role-based access, logs, quality evaluation and GPU monitoring
Measured outcomes
Engineer’s comment
Quality came from more than the model: document chunking rules, permission filtering and a benchmark set that every release must pass.
Customer testimonial
The assistant does not replace engineering judgement, but takes specialists directly to the relevant source passage while keeping data inside our environment.
Need a project with similar logic?