CommercReady: A Technology Transfer n8n Workflow

Secure, self-hosted pipeline for document redaction, multi-agent evaluation, and commercialization strategy.

Self-hosted document intake, multilingual PII redaction, and downstream assessment workflow built with n8n, Microsoft Presidio, and isolated n8n task runners.

The stack is designed for handling confidential files locally. Documents stay inside the mounted confidential_volume directory, while Presidio performs English, Chinese, and mixed-language redaction through internal container-to-container calls orchestrated by a reusable child workflow.

Important Disclaimer

• This workflow does not guarantee that generated analysis, summaries, redaction outcomes, or report content are accurate, complete, or legally sufficient.
• All outputs must be reviewed and validated by qualified human reviewers before any operational, legal, regulatory, or investment decision is made.

License

This repository is licensed under the Apache License 2.0. See LICENSE for details.

Why This Project

• Designed for university technology transfer and research commercialization workflows
• Keeps confidential source documents local, while agents operate on redacted outputs
• Combines deterministic scoring logic with AI synthesis for better auditability
• Supports selective services (Project Evaluation, Market Research, Prior Arts) and full-run strategy mode

Quick Start Path

1. Complete Setup and launch the stack with Docker Compose.
2. Import both workflow JSON files in Import Workflow to n8n.
3. Configure credentials in Configure n8n Credentials.
4. Review Workflow Usage for folder structure and operator flow.
5. Read Evaluation Logic and the detailed Workflow_Logic.md.

Functionality Overview

This project is designed as a practical research-commercialization engine for technology transfer offices, incubators, and university research teams. It turns a confidential project submission into a structured evaluation package without sending raw source files directly into downstream AI analysis.

At a high level, the workflow:

• ingests project metadata and uploaded supporting files from a submission form
• lets operators selectively request project evaluation, market research, and prior arts analysis
• treats strategy as a full-run option that executes the upstream analysis required for final synthesis
• separates and preserves confidential originals inside a controlled local volume
• processes images, office documents, and text-based PDFs into machine-readable content
• flags scanned PDFs that need an additional conversion or OCR step
• redacts personally identifiable information in English, Chinese, and mixed-language documents before any AI assessment step
• builds an in-memory RAG layer over redacted content only
• runs parallel AI agents for project evaluation, market research, prior arts analysis, and commercialization strategy
• computes a deterministic readiness outcome from structured scoring logic rather than relying on freeform model judgment alone
• produces reviewer-ready deliverables such as assessment reports, market research reports, prior arts reports, and strategy plans

The result is a workflow that is useful not only for automation, but also for governance. It keeps raw documents compartmentalized, separates deterministic scoring from LLM synthesis, and produces outputs that are easier to audit, review, and hand off to internal decision-makers.

Sample Data Disclosure

The repository includes a disclosed sample folder at sample_UrbanAir_AI for demonstration and testing.

Important disclosure:

• the "Supporting documents" in this sample set are fabricated content generated using Grok 4.3
• they are provided for workflow demonstration only and should not be treated as real company records, legal filings, or factual evidence

How to use the sample:

• users can review the sample outputs to see both redaction capabilities and generated HTML reports
• the sample outputs were generated by submitting all supporting documents to the workflow
• users can also submit only a few documents to observe how redaction behavior and report content change with partial inputs

Workflow Overview

The image below shows the end-to-end n8n workflow canvas at a high level, including intake, document processing, redaction, RAG construction, parallel service branches, scoring, and strategy generation.

Service selection works as follows:

• Project Evaluation, Market Research, and Prior Arts can be requested independently
• Strategy is not treated as a standalone light-weight branch; when it is selected, the workflow executes the upstream analysis needed to produce the final strategy output

Evaluation Logic

The project evaluation flow is based on a multi-stage business process described in Workflow_Logic.md. That document explains the end-to-end workflow in more detail, including:

• submission intake and folder creation
• multi-format document processing and PII redaction
• in-memory RAG construction over redacted files
• parallel evaluation, market research, and prior arts branches
• weighted scoring, translation-risk penalties, and recommendation logic
• strategy synthesis and report generation

If you want to understand how the scoring and branching decisions work, that document should be treated as the primary reference.

What This Project Includes

• n8n: workflow orchestration and UI
• CommercReady.json: main end-to-end workflow, including language routing and mixed-document orchestration
• CommercReady - Redaction.json: reusable child workflow that performs one Presidio redaction pass for a supplied text and language
• presidio-analyzer: custom Presidio analyzer image with Chinese spaCy support and Hong Kong specific recognizers
• presidio-anonymizer: redaction service used after entity detection
• n8n-runner: external task runner container for isolated code execution
• confidential_volume: local document storage for submissions, redacted outputs, assessments, and strategy reports

Project Structure

.
|-- docker-compose.yml
|-- CommercReady.json
|-- CommercReady - Redaction.json
|-- Dockerfile.n8n
|-- Dockerfile.presidio
|-- Dockerfile.runner
|-- .env.template
|-- config/
|   `-- analyzer.yaml
|-- recognizers/
|   |-- startup.py
|   |-- hk_patterns.py
|   `-- data/
|-- n8n-task-runners.json
`-- confidential_volume/

Setup

Required n8n Community Nodes

This workflow relies on two n8n community nodes. Install both before importing and running the workflow:

• @mazix/n8n-nodes-converter-documents by mazix: https://github.com/mazixs/n8n-node-converter-documents
• n8n-nodes-pdf-page-extract by Oscar Weijman: https://github.com/OscarWeijman/n8n-nodes-pdf-page-extract

These community packages provide the document conversion and PDF page text extraction steps used in the workflow.

1. Install Docker Desktop

This project is intended to run through Docker Desktop on Windows.

1. Go to https://www.docker.com/products/docker-desktop/.
2. Download Docker Desktop for Windows.
3. Run the installer and allow the required Windows features if prompted.
4. Start Docker Desktop and wait until it shows that Docker is running.

Recommended checks:

docker --version

If Docker Desktop starts correctly, that command should return the installed Docker version.

2. Verify Docker Compose

On current Docker Desktop releases, Docker Compose is usually included as the docker compose plugin. In most cases you do not install it separately.

Check that it is available:

docker compose version

If that command fails:

1. Update Docker Desktop to the latest version.
2. Reopen the terminal after Docker Desktop finishes updating.
3. Run docker compose version again.

This repository uses docker compose, not the older docker-compose binary.

3. Prepare the Environment File

Copy the template file:

Copy-Item .env.template .env

Then edit .env and set the values you need. At minimum, review these entries:

• N8N_EDITOR_BASE_URL: base URL for your local n8n editor, normally http://localhost:5678
• N8N_RUNNERS_AUTH_TOKEN: shared secret between n8n and n8n-runner
• GENERIC_TIMEZONE: default timezone for schedules and timestamps
• PRESIDIO_ANALYZER_URL: internal analyzer service URL
• PRESIDIO_ANONYMIZER_URL: internal anonymizer service URL

4. Review the Local Data Mount

The repository mounts ./confidential_volume into the containers at /data/confidential.

This is the main working directory for sensitive project files. Keep in mind:

• input and generated outputs live on your local machine
• containers read and write directly to this folder
• the n8n configuration restricts file access to this path

5. Build and Start the Stack

From the repository root, run:

docker compose up --build

Notes:

• the first build can take longer because the custom Presidio image downloads the Chinese spaCy model
• the n8n editor is exposed on http://localhost:5678
• Presidio analyzer is exposed on http://localhost:5002
• Presidio anonymizer is exposed on http://localhost:5001

To stop the stack:

docker compose down

To stop the stack and remove the named volume used for n8n state:

docker compose down -v

6. Import Workflow to n8n

This project uses two workflow definitions that should be imported together:

• CommercReady.json: the main parent workflow that handles intake, routing, branching, and report generation
• CommercReady - Redaction.json: the reusable child workflow that performs one Presidio redaction pass for a supplied text and language

Important note:

• each user should import their own local copy of both workflows instead of reusing an exported workflow reference from another machine
• n8n Execute Workflow nodes can retain the referenced sub-workflow ID from the exporter’s instance, so the parent workflow must be opened after import and pointed to the child workflow imported in the current n8n instance

Import steps in n8n:

1. Open the n8n editor (http://localhost:5678).
2. Create a new workflow.
3. Use Import from File.
4. Select CommercReady - Redaction.json from the repository root and save it.
5. Create or open another workflow.
6. Use Import from File again.
7. Select CommercReady.json from the repository root and save it.
8. Open the parent workflow and confirm its redaction node calls the child workflow by the imported workflow name.
9. Save both workflows again after confirming the relationship.

When you configure the parent workflow, choose the child workflow from your own local workflow list in n8n. Do not leave a reference to the exporter’s workflow record, because imported Execute Workflow nodes can keep the original sub-workflow ID until you reselect the child workflow in your instance.

After importing, continue with credential setup before running tests or production submissions.

7. Configure n8n Credentials

After the stack is running, open the n8n editor and create the credentials required by the workflow.

Important note:

• red sticky notes on the workflow canvas are placed next to the nodes that require credentials
• make sure each marked node is connected to the correct n8n credential type, because several nodes use similar providers through different node types

Example marker on the workflow canvas:

Credential 1: openRouterApi for OpenRouter account

This credential is used by all LangChain chat-model nodes that power the main AI agents.

Create it in n8n:

1. Go to Credentials.
2. Select New.
3. Choose OpenRouter API.
4. Paste your OpenRouter API key from https://openrouter.ai/keys.
5. Name it something like OpenRouter account.

Nodes that should use this credential:

• Technology Readiness Chat Model: powers the Tech Readiness Evaluation Agent for TRL, IPRL, TMRL, and FRL scoring
• Business Evaluation Chat Model: powers the Business Evaluation Agent for CRL, BRL, and MRL scoring
• Regulatory & Certification Chat Model: powers the Regulatory & Certification Strategy Agent for RRL scoring
• Market Research Chat Model: powers the Market Research Agent
• Tech Describer Chat Model: powers the Tech Describer Agent
• Prior Arts Chat Model: powers the Prior Arts Analysis Agent
• Strategy Chat Model: powers the Strategy Agent

Credential 2: openAiApi for OpenAI account

This credential is used by the embeddings node only.

Create it in n8n:

1. Go to Credentials.
2. Select New.
3. Choose OpenAI API.
4. Fill in your API key and, if needed, a custom base URL.
5. Name it something like OpenAI account.

Node that should use this credential:

• In-memory RAG Embeddings: generates vector embeddings for the redacted document corpus used by the in-memory RAG store

This credential can point to:

• OpenAI directly, such as api.openai.com
• an OpenRouter OpenAI-compatible embeddings endpoint
• Azure OpenAI
• another OpenAI-compatible endpoint

The actual embeddings model is selected in the node configuration, not in the credential itself.

Credential 3: httpBearerAuth for Bearer Auth account

This credential is used by raw HTTP request nodes that call OpenRouter directly instead of using a LangChain chat-model node.

Create it in n8n:

1. Go to Credentials.
2. Select New.
3. Choose Bearer Auth.
4. Paste your OpenRouter API key as the bearer token.
5. Name it something like Bearer Auth account.

Nodes that should use this credential:

• LLM Image Analysis: calls a multimodal OpenRouter model for image description, visible text extraction, and figure analysis
• Web Search Tool: calls an OpenRouter-backed web-search model used by the Market Research Agent

Even though this uses the Bearer Auth credential type, the token value should still be your OpenRouter API key. These nodes send an Authorization: Bearer <key> header, which is what OpenRouter expects.

Credential Summary

Credential name example	n8n credential type	Used by	Purpose
OpenRouter account	openRouterApi	7 nodes	LangChain chat models for all main AI agents
OpenAI account	openAiApi	1 node	Embeddings for the in-memory RAG vector store
Bearer Auth account	httpBearerAuth	2 nodes	Raw HTTP calls to OpenRouter for vision and web search

Nodes That Do Not Need Credentials

These nodes may look like external integrations, but they do not require separate n8n credentials in the current workflow:

• Presidio Anaylze and Presidio Anonymize inside CommercReady - Redaction: these call your local Presidio services and do not have auth configured by default
• Google Patent HTTP Tool: uses a public endpoint and does not require auth in the current setup
• Read/Write File nodes: these operate on the local filesystem
• Form Trigger and Wait nodes: these are native n8n workflow nodes

Workflow Usage

How the Working Folder Is Organized

The main operational data lives inside confidential_volume.

At the top level:

• confidential_volume/: root local storage mounted into the containers as /data/confidential
• confidential_volume/demo.txt: simple sample file for quick testing or mount verification
• confidential_volume/<submission_id>_<project_name>/: one complete working folder for a single submission

Each submission folder acts as the full lifecycle record for one technology transfer case. The folder name is intended to combine a unique submission identifier with a project label so each case is easy to trace and keep separate.

This pattern makes it easier to separate submissions, track processing order, and keep all generated outputs grouped with the original source material.

What Each Submission Folder Is For

Each submission folder is divided into numbered stages.

00_Admin

Administrative and intake metadata for the submission.

Typical contents:

• submission_info.json: structured intake metadata such as project title, category, timestamp, inventor details, consent flag, and submitted file list
• technology_summarization.txt: plain-language technology summary generated by the Tech Describer / Generate Summarization flow and stored in 00_Admin when the selected service includes Strategy or Prior Arts; used by university staff for prior-art searching (for example, CAS SciFinder Prior Art)
• scanned_pdf.json: generated only when uploaded PDF(s) are scanned and text cannot be extracted directly; lists the PDF files that require conversion or OCR handling

Use this folder for:

• storing intake metadata
• preserving workflow audit context
• saving service-dependent project summaries generated early in the process
• flagging scanned PDFs that need a separate conversion step before downstream analysis

01_Original

Original uploaded files before redaction or downstream transformation.

Use this folder for:

• source .docx, .pdf, images, and other raw submission files
• preserving the exact unmodified inputs
• keeping a clean separation between confidential originals and later outputs

This folder should be treated as the most sensitive content area in the workflow.

02_Redacted

Redacted or transformed versions of the original documents.

Typical contents include:

• redacted markdown exports of business plans, team profiles, regulatory documents, and technical descriptions
• image analysis notes generated after processing uploaded figures or scans

Use this folder for:

• storing documents after PII detection and anonymization
• generating safer working copies for internal review
• preparing material for assessment steps that should not depend on raw personal data

03_Assessment

Analytical outputs used to evaluate the submission.

Typical contents include:

• market research reports
• prior art analysis reports
• category-specific assessment reports

Use this folder for:

• storing evaluation artifacts generated from the redacted documents
• collecting decision-support reports for reviewers
• separating analysis deliverables from strategy recommendations

04_Strategy

Final strategy or action-planning outputs derived from the assessment stage.

Typical contents include:

• strategy plan HTML reports
• commercialization or transfer recommendations
• next-step decision documents for internal stakeholders

Use this folder for:

• storing the highest-level recommendation outputs
• preparing handoff material for business or technology transfer review
• keeping strategic conclusions distinct from raw analysis reports

Recommended End-to-End Flow

The folder numbering reflects the intended processing order:

1. Intake metadata and summary are created in 00_Admin.
2. Uploaded confidential source files are stored in 01_Original.
3. Presidio-based redaction and transformation outputs are written to 02_Redacted.
4. Assessment workflows generate reports into 03_Assessment.
5. Final recommendation and planning outputs are written to 04_Strategy.

This structure keeps each submission traceable from intake to final recommendation while preserving a clear boundary between original confidential material and derived working outputs.

Prior Arts Excel Submission

If the workflow includes Prior Arts directly, or if Strategy is selected and therefore triggers the full upstream run, the process will pause at the wait node for prior-arts submission.

Important operator notes:

• the wait node does not automatically pop up the submission form
• open the wait node inside n8n and click the Webform link shown on the right-side panel
• create a new Excel file for the patent list
• place all patent publication numbers in column A only
• enter one publication number per row
• there is no need to rename the worksheet

Example format:

Customization

Environment Customization

Use .env to control runtime behavior without editing container definitions. Common adjustments include:

• execution retention and pruning settings for n8n
• cookie security and editor URL settings
• external runner authentication and resource behavior
• Presidio endpoint variables used by your workflows

If you change .env, restart the stack:

docker compose up -d --build

Presidio Language and Model Configuration

The file config/analyzer.yaml defines the supported languages and NLP models used by the analyzer service.

Current configuration:

• English: en_core_web_lg
• Chinese: zh_core_web_lg

Change this file if you need to add or remove supported languages or swap model choices.

In the current workflow design, the parent workflow handles language classification and routing, while the child workflow performs a single redaction pass for the requested language. Mixed-language documents are handled by calling the child workflow twice in sequence: first zh, then en.

Custom Recognizers

The recognizers folder contains the custom Presidio logic for project-specific detection behavior.

Important files:

• recognizers/startup.py: boots the analyzer service and registers custom recognizers
• recognizers/hk_patterns.py: custom Hong Kong pattern recognizers
• recognizers/result_filter.py: post-processing and filtering logic
• recognizers/data/: CSV-driven rules and lookup data

Use this area when you need to customize:

• organization-specific keywords
• Hong Kong ID or local format detection
• Chinese name handling
• entity alias normalization or result filtering

Because recognizers is mounted into the analyzer container, code and data changes do not require rebuilding the image. In practice, you should restart the analyzer service after changes so the Python process reloads the updated logic:

docker compose restart presidio-analyzer

n8n Runner Customization

The file n8n-task-runners.json controls the external task runner configuration used by n8n.

This is where you can adjust:

• allowed environment variables
• per-runner command configuration
• JavaScript and Python runner restrictions
• health-check ports

Change this carefully. Restrictive defaults are part of the security model for isolated execution.

Docker Service Customization

Use docker-compose.yml when you need to change:

• published ports
• mounted directories
• resource limits
• service dependencies
• health checks

Use the Dockerfiles when you need to change image contents:

• Dockerfile.presidio: install models, Python packages, or custom analyzer assets
• Dockerfile.n8n: extend the n8n base image
• Dockerfile.runner: change runner image behavior or bundled configuration

After editing Dockerfiles or Compose service definitions, rebuild:

docker compose up --build

Troubleshooting

Docker Compose command is not found

• make sure Docker Desktop is installed and running
• open a new terminal after installation or upgrade
• run docker compose version again

n8n does not open on localhost:5678

• confirm the containers are running with docker compose ps
• inspect startup logs with docker compose logs n8n
• verify that port 5678 is not already in use

Presidio analyzer fails on first startup

The first startup may be slow because the Chinese model is loaded during image build and container initialization. If the container exits, inspect logs:

docker compose logs presidio-analyzer

Security Notes

• confidential documents are mounted from the local confidential_volume directory
• file access inside n8n is restricted to /data/confidential
• code execution is isolated through external task runners
• custom recognizers and local data should be reviewed before sharing this repository or its outputs

Credits and References

The evaluation design described in this repository builds on published frameworks and upstream tools rather than inventing its assessment model from scratch.

Methodology references:

• KTH Innovation Readiness Level (IRL) by KTH Innovation / KTH Royal Institute of Technology: source for the core readiness dimensions such as TRL, IPRL, TMRL, FRL, CRL, and BRL
• TT&E Framework by Sebastian (2026): source for Market Readiness Level and the translation-risk concept used to compare technology maturity against market maturity
• Kobos Technology Roadmap (TRM) framework by Kobos et al.: source for the regulatory traction and policy-support perspective used in regulatory readiness analysis

Platform and implementation credits:

• n8n: workflow orchestration and form-driven automation
• Microsoft Presidio: PII detection and anonymization foundation for the multilingual redaction pipeline
• spaCy: NLP model runtime used by the custom Presidio analyzer configuration
• @mazix/n8n-nodes-converter-documents by mazix: https://github.com/mazixs/n8n-node-converter-documents
• n8n-nodes-pdf-page-extract by Oscar Weijman: https://github.com/OscarWeijman/n8n-nodes-pdf-page-extract

Detailed workflow reasoning and methodology attribution are documented in Workflow_Logic.md.