AI Agent n8n: Auto-Score Tender Proposals

This is where everything begins. The n8n Form Trigger node creates a public web form where vendors can submit their proposals directly. When a vendor accesses the form URL, they see a clean interface asking them to upload their proposal document—no login required, no complicated steps.

The form is configured specifically for PDF uploads only, ensuring you receive proposals in a consistent format that the workflow can process. Once a vendor submits their document, the workflow triggers automatically, starting the evaluation process without any manual intervention.

💡 Tip: Copy the Production URL (not the Test URL) when you're ready to share with vendors. The test URL only works when your workflow is in test mode.

This node stores the Google Drive file ID of your original RFP document. Instead of hardcoding the ID throughout the workflow, you define it once here. This makes it easy to switch between different RFP projects—just update this single value.

The workflow uses this ID to download and extract text from your RFP, which the AI then uses as the baseline for evaluating incoming proposals. Think of it as telling the system "this is the standard we're evaluating against."

💡 Tip: To find your Google Drive file ID, open your RFP document in Google Drive and look at the URL. The ID is the long string between /d/ and /view or /edit.

Now the workflow fetches your original RFP document from Google Drive. This node uses the document ID you set in the previous step to download the file as binary data, which will then be converted to extractable text.

The Google Drive integration handles all the authentication and file retrieval automatically. You just need to have your credential configured and the document ID ready.

💡 Tip: Make sure your Google Drive credential has read access to the folder containing your RFP document. If you get permission errors, check the sharing settings on the document.

This node converts your RFP PDF into plain text that the AI can analyze. The Extract From File node processes the binary data downloaded from Google Drive and outputs clean, readable text content.

PDF extraction is crucial because the AI needs actual text to compare against vendor proposals. Without this step, the workflow would just have raw file data that can't be processed intelligently.

💡 Tip: This extraction works best with text-based PDFs. If your RFP contains primarily images or scanned documents, you'll need to use OCR capabilities instead.

This node takes the extracted RFP text and stores it in a clearly named field. The AI agent will reference this content later when comparing vendor proposals against your requirements.

By explicitly naming the field tender_document_content, the workflow maintains clear data organization. This makes it easy to understand what data is flowing where, especially when debugging or modifying the workflow.

The workflow has two parallel data streams at this point: one containing the RFP content, and one containing the vendor's submitted proposal. This Merge node combines them so the AI agent can access both documents simultaneously.

The "Combine by Position" mode aligns the data items so that each evaluation has access to both the original requirements and the proposal being evaluated. This is essential for the comparison analysis.

💡 Tip: If you're processing multiple proposals simultaneously, make sure each proposal is matched with the correct RFP content. The position-based merge works perfectly for single-proposal evaluation.

Just like we extracted text from the RFP, this node extracts text from the vendor's submitted proposal PDF. The binary field name matches the form field label from Step 1, converted to an n8n-compatible format.

This extraction enables the AI to read and analyze the actual proposal content, comparing it point-by-point against your RFP requirements.

💡 Tip: The binary field name is automatically generated from your form field label. If you change the label in Step 1, update this field name accordingly.

This node stores the extracted proposal text in a named field, mirroring what we did for the RFP content. Having both documents stored with clear field names makes the AI prompt construction straightforward.

The workflow now has both the RFP requirements and the vendor proposal as accessible text data, ready for the AI evaluation.

This is the core of the automation. The AI Agent node sends both documents to Google Gemini with specific instructions to perform a rigorous, multi-criteria evaluation. The system message establishes the AI's role as a "demanding tender evaluation specialist with ZERO TOLERANCE for non-compliance."

The prompt includes the RFP content dynamically, asking the AI to evaluate the proposal against those specific requirements. The "Require Specific Output Format" toggle ensures the AI returns structured data that can be processed by subsequent nodes.

💡 Tip: The system message significantly impacts evaluation quality. A more demanding tone produces more thorough, critical analysis. Adjust based on your evaluation standards.

This sub-node specifies which Google Gemini model powers the AI evaluation. The Gemini 2.5 Flash model offers excellent performance for document analysis tasks while maintaining fast response times.

The model selection affects both the quality of analysis and the processing speed. Flash models are optimized for quick responses without sacrificing too much analytical depth.

💡 Tip: If you need more thorough analysis and don't mind longer processing times, consider using gemini-2.5-pro instead.

This Output Parser node defines exactly what data structure the AI should return. By providing a JSON example, you're telling Gemini "give me back data that looks like this." This ensures consistent, machine-readable output that can be processed automatically.

The schema includes company details, cost figures, timeline information, individual criterion scores with weights, compliance flags, and qualitative assessments. Every field you need for your evaluation dashboard is defined here.