Blog
Writing
Notes on building AI systems for production — multi-agent architectures, RAG pipelines, infrastructure decisions, and lessons from shipping.
GraphRAGNeo4jcommunity-detectionLeidenknowledge-graphproduction
Community Detection for Global Queries in Enterprise Knowledge Graphs
How I used Leiden community detection and Map-Reduce to make a company-wide knowledge graph answer big-picture questions like 'What's our biggest technical bottleneck?' — not just ticket-level lookups.
GraphRAGNeo4jQdrantRAGknowledge-graphproduction
When Vector Search Isn't Enough: Adding a Knowledge Graph to RAG
Why vector search alone wasn't enough for a 19K-issue knowledge base — and how layering a knowledge graph on top of semantic retrieval gave the LLM the structural context it was missing.
multi-agentRAGGoogle ADKOpenAI Agents SDKproduction
Building a Multi-Agent Contract Risk Analysis System
How I designed a four-phase agent pipeline for enterprise contract review — with iterative feedback loops, legal RAG, and lessons learned shipping it across three different frameworks.