Broader parser compatibility
Many modern libraries treat DOCX, PPTX, and XLSX as their primary Office inputs, while legacy formats require an installed converter or additional dependencies.
Document normalization layer
Prepare legacy Office knowledge for AI and RAG pipelines
Enterprises often have useful knowledge trapped in Word 97-2003 DOC files, PowerPoint PPT decks, and Excel XLS workbooks. CoreNova converts those legacy binaries into modern OOXML packages before downstream systems parse, structure, chunk, embed, and index the content.
Precise positioning: this API performs format modernization. It does not output Markdown, JSON, semantic chunks, embeddings, or a vector index, and it does not operate a RAG knowledge base.
Reference pipeline
DOC / PPT / XLS→CoreNova conversion→DOCX / PPTX / XLSX→document parser→Markdown / JSON / rows→chunks + metadata→embeddings→vector store + RAG
Keeping conversion separate from extraction makes each stage observable: a team can inspect the modern Office result, select a parser appropriate to the content, and retry later stages without converting the source again.
Why normalize first
Consistent intermediate format
A mixed archive becomes three predictable OOXML families before extraction and quality gates.
Better operational isolation
The conversion stage can fail with a stable reason before expensive parsing, model, embedding, or indexing work begins.
Editable review artifact
Teams can retain and manually inspect a modern Office result instead of treating extracted text as the only output.
Amazon Bedrock Knowledge Bases
Amazon Bedrock Knowledge Bases currently lists DOC/DOCX and XLS/XLSX among supported S3 document types. Conversion is therefore optional for those formats, but it can still be useful when a company wants a standardized OOXML archive or a consistent downstream toolchain.
PowerPoint files are not currently listed in that supported-document table. For PPT or PPTX knowledge, use CoreNova to modernize PPT, then use a presentation-aware parser to produce a Bedrock-supported text, Markdown, HTML, CSV, or PDF representation before ingestion. Always verify the current Bedrock supported-format documentation because service capabilities can change.
Do not claim a direct Bedrock path for every format: CoreNova does not automatically upload to a Bedrock data source, start a sync job, create metadata sidecars, or parse PowerPoint into a Bedrock-supported text format.
Docling and Unstructured
Docling describes DOCX, XLSX, and PPTX as default Office Open XML formats and requires LibreOffice for legacy DOC, XLS, and PPT input. A normalization API can remove that LibreOffice dependency from the customer's parsing environment. See the Docling supported-format guide.
Unstructured supports both legacy and modern Office files, but its legacy Word and PowerPoint paths perform conversion before applying the modern-format partitioner. See the Unstructured partitioning guide. Whether to normalize with CoreNova or inside the parser depends on deployment constraints, volume, support boundaries, and the desired intermediate artifact.
Format-specific extraction strategy
| Result | Recommended extraction unit | Useful metadata | Avoid |
|---|---|---|---|
| DOCX | Heading section, paragraph, list, or table | Source file, heading path, section, modified date, access scope | Splitting only by fixed character count when headings are available |
| PPTX | Slide, slide section, or speaker-note unit | Deck title, slide number, slide title, source path, access scope | Combining the entire deck into one embedding |
| XLSX | Selected sheet, table, row group, or range | Workbook, sheet, range, headers, units, date context | Flattening every sheet and formula into an undifferentiated text block |
Suggested AWS workflow
Inventory legacy files and filter by extension and size before sending any content.
Call the CoreNova API from a trusted Lambda, container, batch job, or application server.
Download successful OOXML results into a customer-controlled staging location.
Run the chosen parser and write text, Markdown, HTML, JSON, CSV, or structured elements.
Attach source identity, access-control, business-domain, and retention metadata.
Chunk, embed, index, evaluate retrieval quality, and retain a trace back to the source document.
Evaluation checklist
- Measure conversion success separately from extraction and retrieval quality.
- Use representative complex files, not only simple test documents.
- Verify that access-control metadata follows every extracted chunk.
- Evaluate tables, charts, slide order, headings, formulas, and document language.
- Keep human review in the migration path for legally or operationally critical documents.
- Do not send API keys, presigned download URLs, or customer documents to support email.
AI-related use cases that fit today
- Normalizing old policy and procedure documents before enterprise AI search.
- Preparing historical project decks for slide-aware extraction.
- Modernizing XLS workbooks before converting selected sheets to structured rows.
- Removing LibreOffice or desktop automation from a downstream parser environment.
- Creating a reviewable OOXML checkpoint in a larger RAG ingestion pipeline.
Capabilities that require another stage
- OCR and scanned-document understanding.
- Layout-aware text, table, chart, or image extraction.
- Markdown, HTML, JSON, or RAG-ready chunk output.
- Embeddings, vector storage, retrieval evaluation, and answer generation.
- Direct S3, SharePoint, OneDrive, or Bedrock synchronization.