ClientMind AI Agent

Inspiration

Standard RAG agents often suffer from "Digital Amnesia"—they can search static documents but fail to "remember" who the customer is. In high-stakes e-commerce, every interaction without context of past preferences, size habits, or sentiment history is a missed sales opportunity. We were inspired to build an agent that doesn't just find information but evolves its memory with every conversation, transforming Elasticsearch into a long-term "MemOS" (Memory Operating System).

What it does

ClientMind is an AI sales agent that transforms Elasticsearch into a Dynamic Persona Store . It automatically monitors incoming customer emails via n8n, uses a Dual-Path RAG to synthesize structured order history (via ES|QL) and product knowledge (via ELSER), and generates hyper-personalized response drafts for human approval.

How we built it

We built the "brain" using Elasticsearch Agent Builder , defining a sophisticated reasoning chain. The "nervous system" is an n8n orchestration workflow that connects Gmail, the Agent APIs, and a Next.js admin dashboard. We utilized ELSER (v2) for semantic knowledge retrieval and ES|QL for real-time customer persona reconstruction, ensuring zero-hallucination personalization.

Challenges we ran into

Our biggest hurdle was architecting the Memory Evolution Loop : extracting unstructured insights from emails and performing asynchronous upserts into Elasticsearch without increasing latency. We also focused heavily on optimizing ES|QL pipelines to ensure persona retrieval stays under 100ms even as the interaction history grows.

Accomplishments that we're proud of

We successfully implemented a Closed-Loop Memory System . Unlike traditional bots, ClientMind actually "learns" from admin-approved emails, updating the customer's long-term profile in real-time. We are also proud of our Dual-Path RAG strategy, which achieves a perfect balance between structured data precision and unstructured semantic depth.

What we learned

We learned that Elasticsearch is far more than a search engine; it is a high-performance Analytical Memory Store . Combining vector search (ELSER) with analytical queries (ES|QL) within a single platform is the most efficient way to build truly "agentic" systems that require both identity and knowledge.

What's next for ClientMind AI Agent

Next, we plan to expand ClientMind to support multi-channel memory (WhatsApp, Instagram) and implement proactive sales triggers —using Elasticsearch to predict when a customer might need a restock based on their "evolved" purchase persona.

Built With

• elasticsearch
• elser
• es|ql
• n8n
• next.js
• react
• tailwind
• typescript