Research made usable
Each content type was processed according to its structure while preserving the source relationships users need to trust and explore an answer.
← Back to Projects
Selected case study · Cultural strategy agent
Problem
Solution
01
02
03
04
Value
Ask Collage
Ask Collage is an AI cultural strategy assistant that draws on years of research data to answer insight, marketing, and strategy questions.
FocusResearch synthesis for strategic decisions
ResultMain product-interest driver for clients; 120% of the Q1 new-business goal
Insights existed, but access depended on already knowing where to look.
Years of cultural research lived across posts, PDFs, presentations, webinar transcripts and survey data. Users needed to ask an insight question in natural language and receive a useful synthesis without identifying a specific report or navigating several pages to find it.
Building the intelligence layer.
The product had to do more than find a matching passage. I shaped the ingestion, synthesis, recommendation and agent workflows so an answer could bring together evidence, explain the strategic signal and open a path into the research behind it.
Designed a source-aware RAG pipeline that processes posts and attachments differently to preserve the context each format carries. Built a custom webinar-segmentation algorithm that identifies topic boundaries from semantic similarity, then combined recency-weighted hybrid retrieval and reranking to surface the most relevant research. Validated citations and connected answers to related reports and deep links, keeping the underlying evidence easy to explore.
Conversations kept intact
Webinar transcripts were segmented around changes in topic, creating coherent research units instead of arbitrary text windows.
Relevance with context
Semantic and lexical signals, recency and custom reranking worked together to select material that was both conceptually close and useful now.
A path beyond the answer
Related research connected insight questions to the survey questions, reports, segments and deeplinks that could move the work forward.
Grounded synthesis, not a query box
The underlying system combined source-aware processing, customized webinar segmentation, hybrid retrieval and reranking. Related-content search added relevance filters, deduplication, LLM judging and canonical-term matching so natural-language questions could lead to the right evidence and research destinations.
I connected the workflows through a reusable LLM framework and authenticated FastAPI service. The agent could work across research and structured data, return citations, suggest follow-ups and preserve a practical response experience under a considered token-cost tradeoff.
An agent that turns insights into action.
Ask Collage gave clients a direct way to move from an insight question to grounded synthesis, supporting evidence and the next useful research path. It became the main product-interest driver for clients and helped the business achieve 120% of its Q1 new-business goal.
Experience
Continue exploring