The situation
Complex documents mix paragraphs, tables and images. Finding one clause, or changing one value, usually means opening files and editing by hand.
Block extraction, search and edit platform for documents
A platform that breaks complex Word documents into searchable, editable content blocks, and regenerates them with the original styling intact.
Document intelligence platform
01 · Context
02 · Challenge
What had to be true
Find the right block from a plain-language question, change it safely, and regenerate the document without breaking its formatting.
03 · What we built
The system
- A block extraction pipeline that parses Word documents into structured JSON, keeping tables, paragraphs, images and their styling
- A custom-trained BERT model that identifies content blocks and gives each a unique ID and canonical block name
- One-time ingestion that enriches metadata with LLMs and NER for indexing
- Chatbot search: natural-language questions become semantically enriched queries with metadata filters
- Chatbot edits: LLMs extract the document ID, target field and new value, apply the change to the source and trigger re-indexing
04 Architecture
How it works, step by step.
Step through the system, or let it play.
- A Word document is ingested
- Blocks extracted to JSON with styles
- BERT names and IDs each block
- LLMs and NER enrich the index
- Search and edit by chat
- Regenerate and re-index
05 · Outcome
Teams search and edit large document sets by asking, and regenerated documents keep their original style.
06 · Stack
Next case · RAG · Gen AIAI chatbots with RAG