AI Creative Research Assistant: Multi-Agent Concept Development Platform

Business Context

‍

A major UK media production group develops a high volume of creative concepts across several in-house teams, converting only a small fraction into commissions each year. Much of the work behind every pitch (landscape research, finding comparable existing work, sourcing supporting data and validating structural logic) was manual, competing directly with time for creative development.

‍

‍

The Challenge

‍

Behind every pitch sat a significant amount of manual research:

‍

• Landscape research across the existing market
• Finding comparable existing work
• Sourcing supporting data
• Validating structural logic

‍

This work consumed time that creative teams needed for developing ideas, limiting how many concepts could reach pitch-ready state.

‍

The Passion Labs Approach

‍

Passion Labs built a multi-agent AI platform that gives each team a research assistant backed by specialist agents, rather than relying on one general-purpose model.

The platform combines:

• An LLM-based ingestion pipeline that turns unstructured public and licensed sources into a clean, structured, quality-checked knowledge base.
• Semantic embeddings for meaning-based similarity search.
• A three-tier memory layer that keeps individual, team, and project context persistent across sessions.

‍

The Solution

‍

The resulting platform functions as a dedicated research assistant for every creative team, handling landscape research, comparable discovery, data sourcing and structural validation. It is deployed on securely isolated cloud infrastructure with strict data segregation between teams, so each team's work stays protected.

‍

‍

‍

Impact & ROI

‍

• ~75% reduction in manual research time, freeing senior creative capacity
• Semantic search surfaced comparators keyword search missed, sharpening originality and differentiation
• Hundreds of records extracted and validated from messy, unstructured sources with automated quality scoring
• More concepts reaching pitch-ready state per sprint
• MVP built and piloted across multiple teams in 10 weeks

‍

‍

‍

This implementation shifts creative teams' time from manual research to concept development, without compromising the quality of the groundwork behind each pitch.

‍