Skip to content
Back to Work
Frontiers Market - Client RAG Chatbots logo
Professional Work

Frontiers Market - Client RAG Chatbots

Completed: Present

Built client-specific RAG chatbots with scoped access to each customer’s own data — evaluating Google’s Agent Development Kit, Zep, and custom vector-store memory side by side rather than locking into one architecture.

The Problem

Clients needed chatbots that answered questions grounded in their own data — not a generic model’s training data — with each client’s chatbot scoped to only its own information. There wasn’t yet a clear answer for which agent/memory architecture would hold up best across different client data shapes and volumes.

The Solution

Used Google’s Agent Development Kit (ADK) to build agents with customer-specific data access as the primary approach, while evaluating alternatives in parallel: Zep for conversational memory, some simpler off-the-shelf chat implementations for lower-complexity cases, and a variation on the custom Qdrant-backed vector memory stores used elsewhere in other engagements. The mix was matched to what each client actually needed rather than forcing one architecture across every deployment.

The Impact

Multiple clients now have chatbots grounded in their own data instead of a generic model’s. Running Google ADK, Zep, and custom vector-store approaches side by side built a clearer picture of the real tradeoffs between them — memory persistence, per-client data isolation, and setup cost — that now informs how new client chatbot work gets scoped.

Technologies

Google ADK
Zep
Qdrant
RAG
LLM Agents

Share This Case Study