At eQ Technologic, I designed a three-tier memory system that turned 10–12 stateless agents into context-aware systems with persistent conversational memory. On-demand retrieval and tiered context composition reduced per-turn token consumption by about 80%.
I also designed a LangGraph multi-agent analytics workflow that converts natural-language queries into BI chart configurations, bringing chart creation down to about a minute. Its agents handle tasks such as filter extraction and follow-up detection, with deterministic validation separated from LLM reasoning.
For a cross-system JIRA reasoning agent, I designed and built a ReAct loop using MCP tool calls to reason across two JIRA systems. The hackathon prototype was later adopted into Customer Training Team material. I’ve also deployed on-premise LLM infrastructure with vLLM and integrated automated vulnerability scanning into CI/CD.
