AICOM

Large language models are powerful, but in enterprise settings the main challenge is reliability. Hallucinations, missing evidence, and untraceable answers make standard LLM assistants hard to deploy responsibly for compliance, reporting, documentation, and decision support.

AICOM (Corporate Knowledge Graph Assistant) develops a new class of enterprise assistants that combine large language models with curated corporate knowledge graphs. The goal is to generate answers that are grounded in internal knowledge, traceable to sources, and accompanied by quality and trust indicators.

What we build

Knowledge graph foundation
We integrate heterogeneous corporate knowledge (structured, semi-structured, and unstructured content) into a coherent, evolving knowledge graph. A central focus is on maintaining consistency, handling updates, and resolving conflicts as knowledge evolves.

LLM + knowledge graph via RAG
We connect the knowledge graph to an LLM using retrieval-augmented generation (RAG), so that answers are produced from retrieved evidence rather than from model memory alone. The aim is context-aware, organization-specific answers that are still verifiable.

Trust and transparency by design
We develop methods to assess and expose answer quality (evidence coverage, coherence, and consistency). The assistant does not only answer, it also explains why the answer makes sense and where it comes from.

Why it matters

Enterprises already experiment with chatbots and assistants, but without traceability these systems remain hard to trust in practical workflows. AICOM targets the gap between:

• Conversational AI (chat-like interaction) and
• Auditable knowledge work (answers that can be checked and justified).

This work is crucial for real-world AI adoption in regulated and knowledge-critical settings (legal, financial, healthcare, engineering documentation).

Public-facing demonstrations

The project is designed for real-world impact and communication. Results will be accompanied by a project website and regular updates. Selected demonstrators are planned for public innovation and science spaces.

Get involved

PhD position (coming soon)

A funded PhD position for AICOM will be advertised publicly. If you are interested, please watch /vacancies/ and /news/ on this website.

Industry / public sector pilots

If you are interested in a pilot use case, please send a short note (max. 1 page) describing:

• your scenario and objectives
• available data sources
• constraints (e.g., on-premise / EU-only deployment, compliance requirements)

Media

For expert interviews on trustworthy enterprise AI, LLMs, chatbots, and evidence-based QA, see /media/.

Partner: WeichertMehner (project coordinator)
Academic lead: TU Dresden / ScaDS.AI (Michael Färber)
Start: 01/04/2026 · Duration: 30 months