Problem
Long documents are difficult to search linearly, while ungrounded chat systems can answer fluently without remaining faithful to the source material.
05 / CASE STUDY
Document intelligence · Retrieval system
Talk with Book
A conversational document system that lets users explore PDFs through retrieval, semantic search and grounded natural-language answers.
RoleAI application architect
StatusApplied AI project
TimelineRAG product exploration
01 / CONTEXTThe problem
The problem
behind the product.
Approach
I structured the product around ingestion, chunking, embeddings, retrieval and context-grounded response generation rather than sending whole documents directly to a model.
02 / SYSTEMArchitecture,
Architecture,
not magic.
01PDF ingestion→
02Chunk & embed→
03Vector retrieval→
04Context assembly→
05Grounded answer
Constraints
- Long and varied PDFs
- Retrieval precision
- Context-window limits
- Source grounding
My ownership
- Designed the document ingestion workflow
- Implemented semantic retrieval
- Structured grounded conversational responses
- Connected the AI pipeline to the product interface
03 / EVIDENCEWhat can be
What can be
defended.
Conversational PDF exploration
Semantic rather than keyword retrieval
Context-grounded responses
Reusable document pipeline
What I would improve now.
A production evolution would add citation-level evaluation, hybrid retrieval, reranking and explicit handling for tables, images and document revisions.
RAGVector searchLLMsPDF processingEmbeddings