Jan 30, 2025
GenovaPlus
An internal RAG system that serves as an enterprise company AI platform
Industry
Telecomunications
Scope of work
AI UX, System Architecture Thinking, Product Design
Duration
Iterative
WHAT THIS WAS
An internal RAG-based AI platform for MTN that began as a single chatbot and evolved into a multi-tenant system where organizations can create tenants, manage users, and build specialized AI agents.
I was the sole product designer, responsible for all major UX decisions, system framing, and design direction.
The Shift That Changed Everything
The original design assumed one AI assistant. New requirements introduced: Multiple organizations (tenants) Multiple agents per tenant Agent-specific knowledge and behavior This forced a fundamental reframing: From “a chatbot” to “an AI agent platform.” That reframing drove every design decision that followed.
INSIGHTS
Key Design Decisions (and Why They Matter)
1. Treating Agents as First-Class Entities Instead of extending a single chat UI, I designed the system around agents as ownable, configurable entities with lifecycle, purpose, and boundaries.This clarified the product’s mental model and allowed it to scale without UX fragmentation. 2. Introducing an Agent Gallery As agents increased, discoverability became critical.I introduced an Agent Gallery — a dedicated space to explore available agents by intent, not configuration. This reduced cognitive load and positioned agents as capabilities, not hidden features.
insights
3. Decentralizing Agent Training A centralized knowledge base worked for one bot — it failed at scale. I made the call to move training and knowledge ownership into each agent, ensuring: - Clear responsibility - Higher relevance - Predictable agent behavior This decision aligned system structure with user expectations.
Why This Project Matters
This wasn’t about designing chat interfaces. It was about: - Designing systems, not screens - Making high-impact decisions under evolving constraints - Translating AI complexity into understandable product structure It represents the kind of problems I’m most effective at solving: ambiguous, platform-level challenges where judgment matters more than polish.
