AI · Enterprise Knowledge · On-Premises
TENGWAR — Enterprise RAG Platform
A private AI assistant that answers questions about your company's documents — running entirely on your own infrastructure, with full citations and no cloud dependency.
Role
Founder · Lead Engineer
Domain
AI · Enterprise Knowledge · On-Premises
Stack
.NET 10 · Blazor · MAUI · Microsoft Agent Framework
01 / Context
Problem & Context
Many organizations would benefit from an AI assistant trained on their internal documents, but cannot send proprietary or regulated material to cloud providers. Legal, GDPR, trade-secret and IP-protection requirements rule out third-party language models. TENGWAR is a plug-and-play on-premises platform: documents are processed locally, indexed in a private vector database, and made queryable from Windows, macOS and web clients — with every answer backed by citations that point to specific pages of the source documents.
Responsibilities
- Founded the product and led design, engineering and operations end-to-end
- Designed the document-understanding pipeline that turns raw files into searchable, multilingual, citable knowledge
- Built the multi-agent AI orchestration: specialist agents for document research, web search, data analysis, communication and document generation
- Delivered the cross-platform client experience (Windows desktop, macOS desktop, web fallback) on a single shared interface
- Engineered the enterprise security model: role-based access, security clearance levels, encrypted credentials, audit-grade logging, default-deny tool authorization
- Orchestrated the on-premises deployment so the entire system runs on a single GPU host, with deterministic resource allocation and full observability
Architecture & Stack
- Single-tenant, on-premises deployment on one GPU-equipped host — no cloud LLM provider is contacted
- Five specialist AI agents (Research, Web, Analytics, Communication, Document Generation) routed by a fast keyword path or by an LLM orchestrator for ambiguous queries
- Smart 18-phase ingestion pipeline: layout detection, OCR, translation, smart chunking and multi-modal indexing — automated end-to-end
- Hybrid retrieval combining text and visual search with cross-encoder reranking and quality gates, so answers stay grounded
- Cross-platform clients (Windows, macOS, web) sharing one Blazor interface; real-time streaming of answers and document-processing progress
- Multi-layered security: encrypted per-user credentials, retrieval-time clearance enforcement, default-deny tool policies, audit trail with redaction
Outcomes
- Private AI assistant for organizations that cannot use cloud LLMs — full data sovereignty, no third-party data egress
- Smart 18-phase ingestion automates layout detection, OCR, translation and indexing across 100+ languages
- Hybrid retrieval delivers grounded, citation-backed answers — every claim traceable to a source page
- Agentic AI autonomously decides when to retrieve documents, search the web, run analysis or draft a deliverable
- Single-GPU footprint makes the platform viable for SMB and government deployments without dedicated AI infrastructure
- Unified Windows, macOS and web clients with a shared experience reduce onboarding friction across mixed teams
Demo
Learn More
Official product page: tengwar.net ↗