A RAG-powered book discovery experience that turns a curated podcast reading list into focused, goal-led recommendations.
Interface studyA conversational interface grounded in a curated reading library.
Overview
A useful system starts with a clear reason to exist.
01 / What was built
A conversational discovery layer for a curated reading library.
02 / Who it was for
Readers arriving with a question, not a title in mind.
03 / Why it mattered
The value in years of source material became easier to reach.
Business challenge
The question was not what technology to use. It was what needed to work better.
The library carried years of useful books and podcast context, but visitors had to browse a static list before they could find the perspective they needed.
Plumfind role / research, strategy, design, engineering, deployment, and support planning.
People were not searching for books.They were searching for answers.
Discovery & research
Reduce uncertainty before production.
01
Business context
The library carried years of useful books and podcast context, but visitors had to browse a static list before they could find the perspective they needed.
02
Research
We mapped the book corpus, podcast themes, user intents, content boundaries, brand tone, and the guardrails needed to keep conversations useful.
03
Experience direction
The experience is intentionally quiet: a small set of goal-led prompts, clear conversational limits, and recommendations that stay close to the source material.
04
Production plan
The application and retrieval services were migrated into the client's AWS environment for ownership, visibility, and long-term control.
Research artefact
Clarity lives in the details.
We mapped the book corpus, podcast themes, user intents, content boundaries, brand tone, and the guardrails needed to keep conversations useful.
Knowledge paths / field note
Reading Room / query
I want to understand how to create a sharper marketing strategy.
Retrieved from the curated library3 relevant reading pathsGrounded in podcast context
Solution
What Plumfind designed and built.
The experience is intentionally quiet: a small set of goal-led prompts, clear conversational limits, and recommendations that stay close to the source material.
01
Structured knowledge base
Book records and podcast context were organized so the system could retrieve relevant source material before responding.
02
Guided conversation
Goal-led prompts, constrained responses, and clear boundaries made the interaction useful without pretending to answer everything.
03
Client-owned production path
The application and retrieval services moved into the client's AWS environment for long-term visibility and ownership.
Architecture
The system is a set of deliberate connections.
Each layer was selected to support the business workflow, not to make the technical picture more complicated.
01Curated sources
02Retrieval index
03Guarded response
04Client environment
01
Curated sources
Books, podcast context, and metadata form the trusted reference layer.
02
Retrieval index
Embeddings make a relevant path through the collection available at query time.
03
Guarded response
The RAG layer keeps answers grounded in retrieved source material.
04
Client environment
AWS services provide an owned operating foundation after launch.
Implementation
From considered direction to a production system.
01
Experience and system design
The experience is intentionally quiet: a small set of goal-led prompts, clear conversational limits, and recommendations that stay close to the source material.
02
Engineering
Book metadata was structured into a retrieval-ready knowledge base, embedded for semantic search, then connected to a guarded RAG response layer.
03
Deployment and ownership
The application and retrieval services were migrated into the client's AWS environment for ownership, visibility, and long-term control.
Business impact
What changed because the system became clearer.
01
Turned a static reading list into an interactive discovery tool
02
Grounded recommendations in a curated knowledge base
03
Kept out-of-scope prompts controlled
04
Transferred deployment ownership to the client
Lessons learned
A good recommendation starts with a well-structured corpus
Guardrails protect relevance and trust
Conversation design is part of the system design
Technologies used
Selected for the work they needed to do.
Technology choices were made around fit, maintainability, and the operating reality of the project.
01
Knowledge layer
RAG, vector embeddings, semantic retrieval
Selected to search meaning and context across the curated library instead of matching only exact words.
02
Response layer
LLMs, guardrails
Selected to make interaction natural while keeping the response aligned to the available source material.
03
Operating foundation
AWS
Selected so the client could own the application, visibility, and future evolution.
