Uploading Documents

Source documents are the foundation of Tendrex. The AI uses these documents - along with your answer library - to generate draft responses to tender questions.

Supported formats

Tendrex accepts the following document formats:

PDF (.pdf)

Including scanned documents with built-in OCR processing via PyMuPDF.

Microsoft Word (.docx)

Standard Word documents with full formatting support.

How to upload

1. Navigate to Source Documents: Select Source Documents from the sidebar in the Tendrex app.
2. Open Upload Dialog: Click the Upload button in the top right corner of the document list.
3. Select Files: Drag and drop your files into the upload area, or click to browse your local file system. You can upload multiple files simultaneously.
4. Start Processing: Click Upload to begin the ingestion pipeline.

Best Practice

For the best AI performance, upload your most recent “best-of” bid responses, company policies, and case studies first.

What happens during ingestion

When a document is uploaded, Tendrex runs a sophisticated ingestion pipeline to prepare it for AI retrieval:

1. Text Extraction: The raw text is extracted. For PDFs, OCR via PyMuPDF ensures even scanned documents are readable.
2. Token-Based Chunking: Text is split into overlapping passages (~500 tokens with 100 token overlap) using tiktoken to preserve context between segments.
3. Vector Embedding: Each chunk is converted into a vector using the built-in all-MiniLM-L6-v2 model (local, no external API calls).
4. Semantic Indexing: Embeddings are stored in ChromaDB, mapped directly to the source document and page numbers for citation traceability.

Note

Processing typically takes 10-30 seconds per document, depending on length and complexity.

Tagging documents

Tags help you organise your knowledge base and control which documents the AI uses when generating answers.

• Add tags during upload or afterwards from the document detail view.
• Filter by tags on the Source Documents page to find specific categories of content.
• Tag-based RAG filtering — when generating answers, the AI can filter source material by tags, ensuring answers are grounded in the most relevant documents.

Tip

Use tags like “policies”, “case-studies”, “previous-bids”, or “technical” to categorise your documents. This makes it easier for the AI to prioritise the right source material for different types of questions.

Managing documents

On the Source Documents page, you have full control over your knowledge base:

• Metadata View: See document names, upload dates, page counts, tags, and real-time processing status.
• Inline Preview: Open the built-in document viewer to see the original file. Word documents are rendered as HTML via Mammoth. PDFs are rendered with full page navigation.
• Library Maintenance: Delete outdated documents to ensure the AI only references current information.

Caution

Deleting a document removes it from the knowledge base. While existing answers that used it remain, the AI will no longer be able to reference it for new questions.

Content strategy

To get the most out of Tendrex, we recommend the following content strategy:

• Upload Master Answers: Your best, most verified responses to common questions.
• Keep it Current: Replace old policies (e.g., GDPR 2018 vs 2024) to avoid the AI providing outdated information.
• Descriptive Naming: Use clear file names (e.g., 2024-ESG-Policy.pdf) to make source citations easier to verify during review.