klbr/n8nworkflows.xyz
1
0
Fork 0
mirror of https://github.com/khoaliber/n8nworkflows.xyz.git synced 2026-04-28 00:29:22 +00:00
Code Issues Packages Projects Releases Wiki Activity

creation

Browse Source
This commit is contained in:
nusquama
2025-11-13 13:23:18 +01:00
parent 34d3bd4af7
commit 09aaee5f92
1 changed files with 301 additions and 0 deletions
Download Patch File Download Diff File Expand all files Collapse all files
workflows/Analyze Legal Documents with GPT & o4-mini plus Gmail & Google Sheets-7610/readme-7610.md
+301
View File
@@ -0,0 +1,301 @@
Analyze Legal Documents with GPT & o4-mini plus Gmail & Google Sheets
https://n8nworkflows.xyz/workflows/analyze-legal-documents-with-gpt---o4-mini-plus-gmail---google-sheets-7610
# Analyze Legal Documents with GPT & o4-mini plus Gmail & Google Sheets
### 1. Workflow Overview
This workflow, titled **"PDF Document Assistant 2.0"**, automates the processing and detailed AI-powered analysis of PDF documents uploaded via a webhook. Its primary use case is to assist users—particularly those handling legal, compliance, or business-critical documents—in understanding complex documents by delivering a professional, richly formatted email report. The workflow integrates PDF text extraction, chunked summarization via OpenAI models, advanced AI document analysis using a LangChain AI Agent, email delivery through Gmail, and logging of interactions in Google Sheets.
The workflow is logically divided into the following functional blocks:
- **1.1 Entry Point & File Reception:** Receives uploaded PDF files and user metadata (name, email, title) via a webhook.
- **1.2 PDF Text Extraction & Preprocessing:** Extracts raw text from the PDF and splits it into manageable chunks enriched with user data.
- **1.3 AI Summarization Chain:** Sequentially summarizes each text chunk using OpenAI's GPT-3.5-turbo model to produce a consolidated summary.
- **1.4 Advanced AI Document Analysis:** Employs a LangChain AI Agent with a custom prompt to perform deep, structured document review and generates a professional email-style analysis.
- **1.5 Email Delivery:** Sends the AI-generated analysis as a formatted email to the user via Gmail.
- **1.6 Logging & Tracking:** Appends metadata about the processed document and user interaction into a Google Sheets spreadsheet for record-keeping and status tracking.
- **1.7 Webhook Response:** Immediately responds to the upload request with a 302 redirect to a success webpage.
---
### 2. Block-by-Block Analysis
#### 2.1 Entry Point & File Reception
- **Overview:**
This block listens for HTTP POST requests on a webhook endpoint, receiving uploaded PDF files alongside user information (name, email, and document title). It provides immediate acknowledgment to the client.
- **Nodes Involved:**
- `Webhook`
- `Respond to Webhook`
- `Sticky Note1`
- `Sticky Note5`
- **Node Details:**
- **Webhook**
- *Type:* HTTP Webhook Listener
- *Configuration:*
Listens on path `/upload-pqx92oa`, HTTP method POST. Returns response via downstream node.
- *Inputs:* External HTTP POST with PDF binary and user metadata.
- *Outputs:* Passes received data (binary PDF, JSON body) downstream.
- *Edge Cases:* Missing or malformed uploads, unsupported file types, large files causing timeouts.
- **Respond to Webhook**
- *Type:* Webhook Response Node
- *Configuration:*
Sends HTTP 302 redirect with "Content-Type: text/html" header to a success page URL upon upload completion.
- *Inputs:* Triggered immediately after file reception.
- *Outputs:* Sends HTTP response to client.
#### 2.2 PDF Text Extraction & Preprocessing
- **Overview:**
Extracts text content from the uploaded PDF. Splits the extracted text into chunks (~4000 characters each) to ensure compatibility with downstream AI model input size limits. Enriches each chunk with user metadata.
- **Nodes Involved:**
- `Extract from File`
- `Code`
- `Sticky Note2`
- **Node Details:**
- **Extract from File**
- *Type:* File Text Extraction
- *Configuration:*
Extracts text from the binary property named `data` using PDF operation.
- *Inputs:* PDF binary from Webhook node.
- *Outputs:* Extracted raw text.
- *Edge Cases:* Corrupted PDFs, encrypted or scanned PDFs without embedded text.
- **Code**
- *Type:* JavaScript Function
- *Configuration:*
Splits extracted text into chunks of 4000 characters; pulls user name, email, and document title from webhook data; outputs array of JSON objects, each with a text chunk and metadata.
- *Inputs:* Extracted text JSON.
- *Outputs:* Array of chunked text objects.
- *Edge Cases:* Empty or very short text, missing user metadata.
#### 2.3 AI Summarization Chain
- **Overview:**
Performs the first level of AI processing: summarizing each chunk of extracted text using OpenAI GPT-3.5-turbo to produce condensed, coherent summaries retaining legal and business meaning.
- **Nodes Involved:**
- `Basic LLM Chain`
- `OpenAI Chat Model1`
- `Code1`
- `Sticky Note3`
- `Sticky Note4`
- **Node Details:**
- **Basic LLM Chain**
- *Type:* LangChain LLM Chain Node
- *Configuration:*
Uses prompt: "Summarize the following section of a document. Retain legal terms, obligations, or business meaning." with input variable `{{ $json.chunk }}`.
- *Inputs:* Chunked text JSON from Code node.
- *Outputs:* Summary text for each chunk.
- **OpenAI Chat Model1**
- *Type:* OpenAI GPT-3.5 Chat Model
- *Configuration:*
Model set to `gpt-3.5-turbo`. Connected as language model for Basic LLM Chain.
- *Credentials:* Uses OpenAI API key (OpenAi account 2).
- *Edge Cases:* Rate limits, API errors, partial responses.
- **Code1**
- *Type:* JavaScript Function
- *Configuration:*
Concatenates all chunk summaries into a single full summary string and attaches user metadata for downstream use.
- *Inputs:* Summarized chunk outputs.
- *Outputs:* Single JSON with `full_summary`, `name`, `email`, and `title`.
#### 2.4 Advanced AI Document Analysis
- **Overview:**
Applies an advanced AI Agent powered by LangChain and OpenAI's o4-mini model to deeply analyze the full summarized document text. It performs classification, detailed chapter-by-chapter review, legal or business insights, risk flagging, and generates a professional email-style report.
- **Nodes Involved:**
- `AI Agent`
- `OpenAI Chat Model`
- `Sticky Note`
- `Sticky Note4`
- **Node Details:**
- **AI Agent**
- *Type:* LangChain Agent Node
- *Configuration:*
Receives a rich prompt designed to guide the AI through expert-level document classification, analysis, and report generation. Uses variables for user name, email, extracted text summary, and document title.
Output is a fully formatted HTML email content.
- *Inputs:* Full summarized text and metadata from Code1.
- *Outputs:* Formatted AI-generated email content.
- *Edge Cases:* Prompt failures, incomplete AI responses, model latency or API errors.
- **OpenAI Chat Model**
- *Type:* OpenAI Chat Model
- *Configuration:*
Model set to `o4-mini-2025-04-16` for deeper reasoning and analysis. Connected as the language model for AI Agent.
- *Credentials:* Uses OpenAI API key (OpenAi account 2).
#### 2.5 Email Delivery
- **Overview:**
Sends the AI Agent's formatted email response directly to the user’s email address, with a subject referencing the document title.
- **Nodes Involved:**
- `Gmail1`
- `Sticky Note6`
- **Node Details:**
- **Gmail1**
- *Type:* Gmail Node
- *Configuration:*
Sends to email address extracted from the original webhook input. Uses subject line “Summary of your attached Document - [Document Title]”.
Message body is the AI Agent's HTML output.
Sender name set to "Swot AI".
- *Credentials:* Uses Gmail OAuth2 credentials ("Gmail account 3").
- *Edge Cases:* SMTP authentication failure, invalid recipient email, Gmail API rate limits.
#### 2.6 Logging & Tracking
- **Overview:**
Logs the processing event, including user name, email, document title, status label (from Gmail node metadata), and timestamp into a Google Sheets document for audit and tracking.
- **Nodes Involved:**
- `Google Sheets`
- **Node Details:**
- **Google Sheets**
- *Type:* Google Sheets Node
- *Configuration:*
Appends a new row with columns: Date, Time, Name, Email, Status, Filename. Uses ISO date/time formatting for timestamp.
Targets specific spreadsheet and sheet (`SwotAI_Users` sheet in a known Google Sheet ID).
- *Credentials:* Google OAuth2 credentials ("Google Sheets account").
- *Edge Cases:* Spreadsheet access permissions, API quota limits, data formatting issues.
---
### 3. Summary Table
| Node Name | Node Type | Functional Role | Input Node(s) | Output Node(s) | Sticky Note |
|-------------------|----------------------------------|--------------------------------------|------------------------|---------------------------|------------------------------------------------------------------------------------------------|
| Webhook | HTTP Webhook | Entry point, receive upload & metadata | - | Extract from File, Respond to Webhook | ## Entry point — receives uploaded PDF via POST request. |
| Respond to Webhook | Webhook Response | Sends immediate HTTP redirect response | Webhook | - | ## Respond to Webhook — sends immediate acknowledgment back to the client after upload. |
| Extract from File | Extract Text from PDF | Extracts raw text from uploaded PDF | Webhook | Code | |
| Code | JavaScript Function | Splits text into chunks, enriches with metadata | Extract from File | Basic LLM Chain | ## Code (pre-processing) - Clean/format the extracted text (remove line breaks, non-text). |
| Basic LLM Chain | LangChain LLM Chain | Summarizes each text chunk | Code | Code1 | ## Basic LLM Chain - First AI pass: summarization / restructuring of document content. |
| OpenAI Chat Model1 | OpenAI GPT-3.5 Chat Model | Provides language model for summarization | Basic LLM Chain (ai_languageModel) | Basic LLM Chain | |
| Code1 | JavaScript Function | Combines chunk summaries and metadata | Basic LLM Chain | AI Agent | |
| AI Agent | LangChain Agent Node | Deep document analysis & report generation | Code1 | Gmail1 | ## AI Agent - Central intelligence handling prompts, memory, reasoning, and detailed analysis. |
| OpenAI Chat Model | OpenAI Chat Model (o4-mini) | Language model for AI Agent | AI Agent (ai_languageModel) | AI Agent | |
| Gmail1 | Gmail Node | Sends formatted analysis email to user | AI Agent | Google Sheets | ## Gmail - Sends formatted results to the user (summary, clauses, flagged risks, etc.). |
| Google Sheets | Google Sheets Append Row | Logs user and document metadata | Gmail1 | - | |
| Sticky Note1 | Sticky Note | Annotation for Webhook block | - | - | ## Entry point — receives uploaded PDF via POST request. |
| Sticky Note2 | Sticky Note | Annotation for Code pre-processing | - | - | ## Code (pre-processing) - Clean/format the extracted text (remove line breaks, non-text). |
| Sticky Note3 | Sticky Note | Annotation for Basic LLM Chain | - | - | ## Basic LLM Chain - First AI pass: summarization / restructuring of document content. |
| Sticky Note4 | Sticky Note | Annotation for AI Agent | - | - | ## AI Agent - Central intelligence handling prompts, memory, reasoning, and detailed analysis. |
| Sticky Note5 | Sticky Note | Annotation for Respond to Webhook | - | - | ## Respond to Webhook - Sends immediate acknowledgment back to the client after upload. |
| Sticky Note6 | Sticky Note | Annotation for Gmail node | - | - | ## Gmail - Sends formatted results to the user (summary, clauses, flagged risks, etc.). |
---
### 4. Reproducing the Workflow from Scratch
1. **Create Webhook Node**
- Type: HTTP Webhook
- Parameters:
- Path: `upload-pqx92oa`
- HTTP Method: POST
- Response Mode: Response Node
- Purpose: Receive uploaded PDF and user metadata.
2. **Create Respond to Webhook Node**
- Type: Respond to Webhook
- Parameters:
- Response Code: 302
- Response Headers: Content-Type = text/html
- Redirect URL: `https://swot-ai25.github.io/pdf-document-assistant/success.html`
- Connect Webhook → Respond to Webhook.
3. **Create Extract from File Node**
- Type: Extract from File
- Parameters:
- Operation: pdf
- Binary Property Name: `data` (matches uploaded file property)
- Connect Webhook → Extract from File.
4. **Create Code Node (Text Chunking & Metadata Injection)**
- Type: Code (JavaScript)
- Parameters:
- JavaScript to:
- Retrieve extracted text from `text` property
- Retrieve `name`, `email`, and `title` from webhook JSON body
- Split text into 4000-character chunks
- Output array of chunk objects, each with chunk text and metadata
- Connect Extract from File → Code.
5. **Create Basic LLM Chain Node**
- Type: LangChain Chain LLM
- Parameters:
- Prompt: "Summarize the following section of a document. Retain legal terms, obligations, or business meaning.\n\n{{ $json.chunk }}"
- Connect Code → Basic LLM Chain.
6. **Create OpenAI Chat Model Node (GPT-3.5 Turbo)**
- Type: OpenAI Chat Model
- Model: `gpt-3.5-turbo`
- Assign credentials for OpenAI API.
- Connect as language model to Basic LLM Chain.
7. **Create Code1 Node (Concatenate Summaries)**
- Type: Code (JavaScript)
- Parameters:
- Combine all chunk summaries into a single string in `full_summary`
- Pass along `name`, `email`, and `title` from earlier steps
- Connect Basic LLM Chain → Code1.
8. **Create AI Agent Node**
- Type: LangChain Agent
- Parameters:
- Use detailed prompt template for document classification, chapter-by-chapter analysis, legal/business insights, risk flagging, and professional email generation (HTML format)
- Connect Code1 → AI Agent.
9. **Create OpenAI Chat Model Node (o4-mini)**
- Type: OpenAI Chat Model
- Model: `o4-mini-2025-04-16`
- Assign credentials for OpenAI API.
- Connect as language model to AI Agent.
10. **Create Gmail Node**
- Type: Gmail
- Parameters:
- Send To: `={{ $('Code1').item.json.email }}`
- Subject: `Summary of your attached Document - {{ $('Code1').item.json.title }}`
- Message: `={{ $json.output }}` (HTML formatted from AI Agent)
- Sender Name: "Swot AI"
- Append Attribution: false
- Assign Gmail OAuth2 credentials.
- Connect AI Agent → Gmail.
11. **Create Google Sheets Node**
- Type: Google Sheets Append Row
- Parameters:
- Document ID and Sheet Name for user logs
- Columns: Date (current date), Time (current time), Name, Email, Status (from Gmail labelIds), Filename
- Assign Google Sheets OAuth2 credentials.
- Connect Gmail → Google Sheets.
12. **Add Sticky Notes** for documentation (optional but recommended):
- Mark entry point, code preprocessing, AI summarization, AI Agent, response, and email sending steps.
---
### 5. General Notes & Resources
| Note Content | Context or Link |
|--------------------------------------------------------------------------------------------------------------------------------------------------------------|-------------------------------------------------------------------------------------------------|
| The workflow uses LangChain nodes for advanced AI orchestration, integrating OpenAI chat models for summarization and detailed reasoning. | https://docs.n8n.io/integrations/builtin/nodes/n8n-nodes-langchain/ |
| Gmail node uses OAuth2 for secure email sending; ensure Gmail account is authorized with appropriate scopes (send email). | https://developers.google.com/gmail/api/auth/about-auth |
| Google Sheets node appends metadata logs for audit and tracking; requires Google Sheets API enabled and OAuth2 credentials configured. | https://developers.google.com/sheets/api/quickstart/js |
| The chunk size of 4000 characters balances prompt length limits and summarization quality for GPT-3.5-turbo. | GPT-3.5-turbo max tokens ~4096 tokens; chunk size chosen conservatively for safe prompt limits. |
| The AI Agent prompt is designed for legal/business document analysis with professional email formatting in HTML, suitable for direct user communication. | Prompt text embedded in AI Agent node parameters, following best practices for clarity and tone.|
| Immediate webhook response prevents client timeouts and provides user feedback on upload success via a redirect to a static success page. | https://swot-ai25.github.io/pdf-document-assistant/success.html |
---
This completes a detailed, structured reference of the **"PDF Document Assistant 2.0"** workflow, enabling understanding, reproduction, and extension.