Agent-Driven Category Intelligence Narrative Engine on Wippy
Challenge
Intent AI's AiRICA platform runs on a category-management methodology built over 25 years. Every client engagement required a data-science team to manually reconcile four structurally different datasets: a consumer survey, syndicated POS data, panel-leakage data, and a retailer SKU export. The GPS framework that should have connected every analytical layer existed only in the team's heads. The output was rebuilt in Excel and PowerPoint after every engagement, making the methodology impossible to scale. A second problem was that a general-purpose LLM would prescribe actions at steps where the methodology only called for description, which destroyed credibility with category managers who knew the methodology.
Intent AI's AiRICA platform runs on a category-management methodology built over 25 years. Every client engagement required a data-science team to manually reconcile four structurally different datasets: a consumer survey, syndicated POS data, panel-leakage data, and a retailer SKU export. The GPS framework that should have connected every analytical layer existed only in the team's heads. The output was rebuilt in Excel and PowerPoint after every engagement, making the methodology impossible to scale. A second problem was that a general-purpose LLM would prescribe actions at steps where the methodology only called for description, which destroyed credibility with category managers who knew the methodology.
Solution
A standalone narrative engine runs on Wippy, an open-source actor-model runtime, with its own data store and no dependency on AiRICA's existing backend. An ingestion layer maps each source to a canonical schema and joins on Item ID, treating unmatched SKUs as unmapped. A chain of Wippy agents each owns a specific analytical question and writes findings to shared state. The narrative agent reads that state and produces a connected category growth story. Intent AI's editorial rules, when the system can recommend versus when it must stay descriptive, are encoded as per-step agent prompts from the client's Demo Business Questions library. Those rules sit in configuration the client controls, not in code.
A standalone narrative engine runs on Wippy, an open-source actor-model runtime, with its own data store and no dependency on AiRICA's existing backend. An ingestion layer maps each source to a canonical schema and joins on Item ID, treating unmatched SKUs as unmapped. A chain of Wippy agents each owns a specific analytical question and writes findings to shared state. The narrative agent reads that state and produces a connected category growth story. Intent AI's editorial rules, when the system can recommend versus when it must stay descriptive, are encoded as per-step agent prompts from the client's Demo Business Questions library. Those rules sit in configuration the client controls, not in code.
Results
A narrative engine and visual prototype now exist that take Intent AI's four source datasets, run them through a GPS-anchored agent chain, and produce a connected category growth story with operational implications. The engine ran real mattress retail data in the first production run and now runs a client deliverable on real retail data and is demoed to clients and investors. All deliverables transferred to Intent AI as work-made-for-hire on payment. The Intent AI team refines agent prompts, knowledge base content, and narrative thresholds without filing engineering tickets, an explicit contractual deliverable.
A narrative engine and visual prototype now exist that take Intent AI's four source datasets, run them through a GPS-anchored agent chain, and produce a connected category growth story with operational implications. The engine ran real mattress retail data in the first production run and now runs a client deliverable on real retail data and is demoed to clients and investors. All deliverables transferred to Intent AI as work-made-for-hire on payment. The Intent AI team refines agent prompts, knowledge base content, and narrative thresholds without filing engineering tickets, an explicit contractual deliverable.
