Speaker
Description
Large Language Models are increasingly being embedded into production systems, yet many current Retrieval-Augmented Generation (RAG) approaches struggle with trust, traceability, and operational robustness.
This talk explores how Agentic RAG combined with Knowledge Graphs can be used to build verifiable, auditable knowledge systems rather than opaque AI applications. Using Telos, a recently released legal reasoning engine, as a concrete case study, the session examines how complex domain knowledge can be structured, retrieved, reasoned over, and iteratively refined.
Topics covered include:
- Limitations of naïve RAG architectures in real systems
- Agent-based retrieval and reasoning workflows
- Knowledge graphs as structural infrastructure for grounding and provenance
- Design patterns for inspection, traceability, and controlled generation
- Infrastructure considerations for deploying and operating such systems
Although the case study is drawn from the legal domain, the architectural principles apply broadly to compliance, finance, internal knowledge platforms, and other knowledge-intensive systems.
Justification
This talk will be valuable to UbuCon attendees because it bridges open infrastructure thinking with emerging AI-driven knowledge systems. Many engineers are being asked to deploy LLM-based solutions without clear architectural guidance on reliability, observability, and trust.
The framing of Agentic RAG and Knowledge Graphs as infrastructure components rather than AI features, the session provides attendees with concrete patterns they can apply in open-source, cloud-native, and on-prem environments. The talk emphasizes failure modes, system boundaries, and operational considerations that resonate strongly with platform and infrastructure engineers.
| Technical level | intermediate |
|---|---|
| Submission type | Talk |
| Where are you based? | Nairobi |