Agentic AI, LLM engineering, and the models behind modern automation — multi-agent systems, LLM evaluation and comparisons, RAG, fine-tuning, AI infrastructure, security, and production AI engineering.
From the blog
GraphRAG moved from research to production in 2026. Three implementations side-by-side: Neo4j-based, Microsoft GraphRAG, and Graphiti's temporal graph approach.
GQA and MLA cut KV-cache memory by huge factors. The 2026 implementations and the production tradeoffs that decide which one to use.
Goal trees decompose complex objectives into manageable subgoals. The 2026 patterns for building, traversing, and pruning them in production.
Horizontal scaling for LLM-backed APIs has surprises traditional APIs do not. The 2026 patterns and the pitfalls that bite.
Pure dense retrieval is not enough. The 2026 hybrid search stack that combines BM25, dense, ColBERT-V2, and learned sparse vectors.
Indirect prompt injection is the top agentic-AI vulnerability of 2026. The ten attack vectors actually being exploited in production.
Patterns and pitfalls for putting LLMs inside CRMs in 2026 — auth, audit, data residency, and the integrations that actually pay back.
Late-interaction retrievers like ColPali and Jina-ColBERT changed how RAG works on documents and images in 2026. The architecture and where it wins.
Code-Review-Graph runs Leiden community detection over your codebase to surface architectural communities, hubs, and bridge modules. Here is why that matters for AI-assisted refactoring.
