Meta AI Business Assistant on Instagram | Meta
Project overview
Context: The Meta AI Business Assistant (MAIBA) on Instagram provides autonomous, contextual ad setup and troubleshooting support directly within native business surfaces — including the high-volume Instagram Boost workspace.
The Challenge: Small business owners frequently encounter technical blockers or policy confusion when trying to boost a post. Early development models hit retrieval bottlenecks; fragmented help data caused poor RAG (Retrieval-Augmented Generation) quality, meaning the agent would surface incorrect instructions or fail to resolve localized intent paths during the ad setup flow.
The Goal: Restructure the documentation architecture, author high-impact launch content, and refine intent rules so the model can reliably guide users through successful Instagram Boost configurations.
I owned the structural layout, launch content creation, and model evaluation loops:
Launch Content & UX Mapping: Authored and deployed a completely new suite of structural support and user education content specifically for the Instagram Boost x MAIBA launch, formatting nodes so they could be seamlessly ingested by both the AI model and front-end contextual UI tooltips.
RAG Information Architecture: Cleaned legacy help content, building structured articles that eliminate semantic confusion so the model can reliably pull accurate troubleshooting answers during live merchant chat sessions.
Intent Scaling & Query Development: Created comprehensive user query datasets to document linguistic ambiguities, directly scaling how the bot recognizes and routes complex merchant intents within the Instagram advertising flow.
Ground-Truth Evaluation: Directed multi-dimensional quality evaluations to produce ground-truth labels for retrieval accuracy, conversational voice consistency, and taxonomy precision to train automated LLM judges.
My Role
Product Manager
Product Marketing Manager
Content Designer
Product Engineers
Key Collaborators
Single Source of Truth Node
Improved AI retrieval quality, completely consolidating overlapping data structures into a unified, machine-readable repository.
New Product Entry Integration
Successfully authored and deployed specialized knowledge nodes for the flagship Instagram Boost x MAIBA launch interface, scaling automated entry points for global merchants.