Cogniviva

Cogniviva started with a straightforward frustration: sales teams know their product deeply but can't hold it all in their head at once, and prospects generate questions that no one has cleanly pre-answered. The gap between what a company knows and what it can articulate in a sales call is almost always larger than it looks.

The system maintains two parallel knowledge bases — one for the product (features, positioning, competitor differentiators, pricing logic) and one for each prospect (firmographics, tech stack, buying signals, previous conversations). When a rep needs to pitch, the engine fuses relevant fragments from both bases and routes them through an LLM to generate a coherent, contextualised opening narrative or response to a specific objection.

How it's built

The back-end is Haskell + Yesod, chosen because the query and fusion logic benefits from strong typing and the concurrency model that comes free with GHC. Server-Sent Events push streaming LLM output to the front-end in real time so reps see the answer assemble itself rather than waiting on a spinner.

The front-end is a single-page application built on Cappuccino, an Objective-C-inspired framework that treats the browser like a thick client — rich object graphs, bindings and responder chains rather than the React model. It's an unusual choice in 2024, but it matches the kind of stateful, desktop-like UI that sales tooling actually needs. Data lives in PostgreSQL with a vector extension for semantic retrieval.

Current state

Cogniviva is under active development. The core knowledge management pipeline, semantic search and pitch generation are working; the prospect-enrichment and call-logging integrations are the next phase. It started as an independent project and has since absorbed Zeus, an earlier experiment in product-knowledge graphs, as its ontology layer.