Alvar Knowledge

Overview

Alvar Knowledge is an AI knowledge platform built on the Schmitz-Stiftungen's 40-year, 150 GB archive. It answers questions in natural language and grounds every sentence in a real, checkable source. The platform carries a hard guarantee: citation existence is validated by the system, and any statement it cannot back with a real source is discarded before the user ever sees it. A LangGraph question-answering agent orchestrates retrieval, grounding, and validation across the archive. The result is institutional knowledge that stays trustworthy at scale, with no room for invented references. It is live at the foundation today and was presented at the Deutscher Stiftungstag 2026 in Hamburg.

Specifications

• AGENT — LangGraph question-answering agent

• STACK — Python/FastAPI, PostgreSQL 16, Qdrant, Redis, Celery

• MODELS — Gemini via Vertex AI in the EU region

• SECURITY — PostgreSQL 16 with row-level security

• ANONYMIZATION — Presidio + spaCy with k-anonymity

• PRICING — Tiered, from a small-foundation plan upward

Features

1. Natural-language answers over a 40-year, 150 GB archive

2. Every sentence grounded in a real, checkable source

3. Hard guarantee: system-validated citation existence

4. Unbacked statements discarded before the user sees them

5. LangGraph agent orchestrating retrieval and grounding

6. Row-level security isolating access at the data layer

7. EU-region Gemini models via Vertex AI

8. Anonymization with Presidio + spaCy and k-anonymity

9. Opt-in Common Pool across institutions, opening to the sector in 2027

Architecture

GROUNDED QUESTION-ANSWERING FLOW

1. Question in natural language

2. LangGraph agent retrieves from the 150 GB archive

3. Qdrant vector store surfaces candidate sources

4. Citation existence validated by the system

5. Unbacked statements discarded

6. Grounded, source-checked answer to the user