by Trung Tran
CV Extractor: Google Drive to Sheet + Slack Update for Recruiters Watch the demo video below: > This workflow automatically processes resumes (PDFs) uploaded or updated in a Google Drive folder. It extracts and structures the candidate’s information using AI, then updates or inserts the data into a Google Sheet, acting as a central talent database. Finally, it notifies the hiring team via Slack with a summary. Perfect for HR and TA teams, this automation eliminates the repetitive task of manually copying candidate details from CVs into spreadsheets, saving hours of admin work every week and keeping your hiring pipeline clean, fast, and up to date. 👤 Who’s it for This workflow is designed for: Recruiters* and *HR coordinators** who manage candidate profiles via Google Drive. Talent Acquisition teams** who want to automate CV parsing, enrichment, and database updating. Companies or hiring agencies** using spreadsheets for candidate tracking and CRM-like HR ops. ⚙️ How it works / What it does This smart and fully automated workflow: Monitors a Google Drive folder for any uploaded or updated resumes (PDFs). Downloads and extracts resume content using PDF parsing. Sends the raw text to GPT-4, which returns a structured profile (name, title, experience, skills, etc.). Verifies the profile and transforms it into a clean, row-based format. Upserts the candidate profile into a Google Sheet (insert or update by email). Notifies the hiring team in Slack or email that a profile was added or updated. This is a no-touch pipeline to keep your candidate data clean, current, and centralized. 🛠️ How to set up Step 1: Prepare your Google Drive folder Create a folder like /SmartHR/cv/ Upload sample resumes in .pdf format Step 2: Create your Google Sheet Columns to include: Email, FullName, JobTitle, Phone, Location, Experience, Education, Skills, etc. Optional: Add conditional formatting to highlight updates Step 3: Connect the n8n workflow Use the Google Drive Trigger: fileCreated → new profile uploaded fileUpdated → existing profile modified Use Google Drive (Download file) to fetch the resume Use Extract From PDF to get raw content Step 4: Configure GPT-4 node Use the structured system prompt to extract profile information Use json parser node to ensure safe formatting for next steps Step 5: Transform & Save Use a Function node to map fields to Google Sheet columns Use Append or update row (based on email as unique key) Optionally send Slack or email message to notify hiring team ✅ Requirements 🔑 OpenAI GPT-4 API key 🟩 n8n Cloud or Self-hosted with: Google Drive integration Google Sheets integration Email/Slack credentials (optional) 📄 Resume files in readable PDF format 📊 Google Sheet prepared with relevant headers ✏️ How to customize the workflow | Part | Customization Options | |----------------------------|----------------------------------------------------------------------------------------| | GPT Prompt | Tune for different job levels or fields (e.g., engineers vs marketers) | | Field Mapping | Update transform node to include other profile fields (LinkedIn, portfolio, etc.) | | Notification | Switch to Microsoft Teams, Telegram, or email alerts instead of Slack | | Data Store | Replace Google Sheet with Airtable, Notion, or database system | | Trigger Source | Trigger from email attachments or webhook instead of Google Drive if needed | | Output Format | Generate PDF profile cards or summary documents using HTML → PDF node |
by Yaron Been
Vcollos Trefilio AI Generator Description None Overview This n8n workflow integrates with the Replicate API to use the vcollos/trefilio model. This powerful AI model can generate high-quality other content based on your inputs. Features Easy integration with Replicate API Automated status checking and result retrieval Support for all model parameters Error handling and retry logic Clean output formatting Parameters Required Parameters prompt** (string): Prompt for generated image. If you include the trigger_word used in the training process you are more likely to activate the trained object, style, or concept in the resulting image. Optional Parameters mask** (string, default: None): Image mask for image inpainting mode. If provided, aspect_ratio, width, and height inputs are ignored. seed** (integer, default: None): Random seed. Set for reproducible generation image** (string, default: None): Input image for image to image or inpainting mode. If provided, aspect_ratio, width, and height inputs are ignored. model** (string, default: dev): Which model to run inference with. The dev model performs best with around 28 inference steps but the schnell model only needs 4 steps. width** (integer, default: None): Width of generated image. Only works if aspect_ratio is set to custom. Will be rounded to nearest multiple of 16. Incompatible with fast generation height** (integer, default: None): Height of generated image. Only works if aspect_ratio is set to custom. Will be rounded to nearest multiple of 16. Incompatible with fast generation go_fast** (boolean, default: False): Run faster predictions with model optimized for speed (currently fp8 quantized); disable to run in original bf16 extra_lora** (string, default: None): Load LoRA weights. Supports Replicate models in the format <owner>/<username> or <owner>/<username>/<version>, HuggingFace URLs in the format huggingface.co/<owner>/<model-name>, CivitAI URLs in the format civitai.com/models/<id>[/<model-name>], or arbitrary .safetensors URLs from the Internet. For example, 'fofr/flux-pixar-cars' lora_scale** (number, default: 1): Determines how strongly the main LoRA should be applied. Sane results between 0 and 1 for base inference. For go_fast we apply a 1.5x multiplier to this value; we've generally seen good performance when scaling the base value by that amount. You may still need to experiment to find the best value for your particular lora. megapixels** (string, default: 1): Approximate number of megapixels for generated image How to Use Set up your Replicate API key in the workflow Configure the required parameters for your use case Run the workflow to generate other content Access the generated output from the final node API Reference Model: vcollos/trefilio API Endpoint: https://api.replicate.com/v1/predictions Requirements Replicate API key n8n instance Basic understanding of other generation parameters
by Viktor Klepikovskyi
Preventing Google Sheets Quota Errors during Batch Processing This template provides a robust solution for dealing with Google Sheets API rate limits. It is designed for workflows that update a large number of rows in a Google Sheet and frequently fail with "too many requests" errors. The template uses a Wait node connected to the error output of the Google Sheets node, creating a retry loop that delays execution for a set period before attempting the update again. To use this template, simply replace the placeholder Google Sheets nodes with your own credentials and sheet. You can find an example Google Sheet for this template here. For a full explanation of this approach, check out the blog post here.
by Yaron Been
Izzaanel Betia AI Generator Description None Overview This n8n workflow integrates with the Replicate API to use the izzaanel/betia model. This powerful AI model can generate high-quality other content based on your inputs. Features Easy integration with Replicate API Automated status checking and result retrieval Support for all model parameters Error handling and retry logic Clean output formatting Parameters Required Parameters prompt** (string): Prompt for generated image. If you include the trigger_word used in the training process you are more likely to activate the trained object, style, or concept in the resulting image. Optional Parameters mask** (string, default: None): Image mask for image inpainting mode. If provided, aspect_ratio, width, and height inputs are ignored. seed** (integer, default: None): Random seed. Set for reproducible generation image** (string, default: None): Input image for image to image or inpainting mode. If provided, aspect_ratio, width, and height inputs are ignored. model** (string, default: dev): Which model to run inference with. The dev model performs best with around 28 inference steps but the schnell model only needs 4 steps. width** (integer, default: None): Width of generated image. Only works if aspect_ratio is set to custom. Will be rounded to nearest multiple of 16. Incompatible with fast generation height** (integer, default: None): Height of generated image. Only works if aspect_ratio is set to custom. Will be rounded to nearest multiple of 16. Incompatible with fast generation go_fast** (boolean, default: False): Run faster predictions with model optimized for speed (currently fp8 quantized); disable to run in original bf16 extra_lora** (string, default: None): Load LoRA weights. Supports Replicate models in the format <owner>/<username> or <owner>/<username>/<version>, HuggingFace URLs in the format huggingface.co/<owner>/<model-name>, CivitAI URLs in the format civitai.com/models/<id>[/<model-name>], or arbitrary .safetensors URLs from the Internet. For example, 'fofr/flux-pixar-cars' lora_scale** (number, default: 1): Determines how strongly the main LoRA should be applied. Sane results between 0 and 1 for base inference. For go_fast we apply a 1.5x multiplier to this value; we've generally seen good performance when scaling the base value by that amount. You may still need to experiment to find the best value for your particular lora. megapixels** (string, default: 1): Approximate number of megapixels for generated image How to Use Set up your Replicate API key in the workflow Configure the required parameters for your use case Run the workflow to generate other content Access the generated output from the final node API Reference Model: izzaanel/betia API Endpoint: https://api.replicate.com/v1/predictions Requirements Replicate API key n8n instance Basic understanding of other generation parameters
by Yaron Been
Justingirard Draft Ui Designer Image Generator Description An experiment: a fine-tuned FLUX model for UI design generation Overview This n8n workflow integrates with the Replicate API to use the justingirard/draft-ui-designer model. This powerful AI model can generate high-quality image content based on your inputs. Features Easy integration with Replicate API Automated status checking and result retrieval Support for all model parameters Error handling and retry logic Clean output formatting Parameters Required Parameters prompt** (string): Prompt for generated image. If you include the trigger_word used in the training process you are more likely to activate the trained object, style, or concept in the resulting image. Optional Parameters mask** (string, default: None): Image mask for image inpainting mode. If provided, aspect_ratio, width, and height inputs are ignored. seed** (integer, default: None): Random seed. Set for reproducible generation image** (string, default: None): Input image for image to image or inpainting mode. If provided, aspect_ratio, width, and height inputs are ignored. model** (string, default: dev): Which model to run inference with. The dev model performs best with around 28 inference steps but the schnell model only needs 4 steps. width** (integer, default: None): Width of generated image. Only works if aspect_ratio is set to custom. Will be rounded to nearest multiple of 16. Incompatible with fast generation height** (integer, default: None): Height of generated image. Only works if aspect_ratio is set to custom. Will be rounded to nearest multiple of 16. Incompatible with fast generation go_fast** (boolean, default: False): Run faster predictions with model optimized for speed (currently fp8 quantized); disable to run in original bf16 extra_lora** (string, default: None): Load LoRA weights. Supports Replicate models in the format <owner>/<username> or <owner>/<username>/<version>, HuggingFace URLs in the format huggingface.co/<owner>/<model-name>, CivitAI URLs in the format civitai.com/models/<id>[/<model-name>], or arbitrary .safetensors URLs from the Internet. For example, 'fofr/flux-pixar-cars' lora_scale** (number, default: 1): Determines how strongly the main LoRA should be applied. Sane results between 0 and 1 for base inference. For go_fast we apply a 1.5x multiplier to this value; we've generally seen good performance when scaling the base value by that amount. You may still need to experiment to find the best value for your particular lora. megapixels** (string, default: 1): Approximate number of megapixels for generated image How to Use Set up your Replicate API key in the workflow Configure the required parameters for your use case Run the workflow to generate image content Access the generated output from the final node API Reference Model: justingirard/draft-ui-designer API Endpoint: https://api.replicate.com/v1/predictions Requirements Replicate API key n8n instance Basic understanding of image generation parameters
by Yaron Been
Settyan Flash V2.0.1 Beta.10 AI Generator Description None Overview This n8n workflow integrates with the Replicate API to use the settyan/flash-v2.0.1-beta.10 model. This powerful AI model can generate high-quality other content based on your inputs. Features Easy integration with Replicate API Automated status checking and result retrieval Support for all model parameters Error handling and retry logic Clean output formatting Parameters Required Parameters prompt** (string): Prompt for generated image. If you include the trigger_word used in the training process you are more likely to activate the trained object, style, or concept in the resulting image. Optional Parameters mask** (string, default: None): Image mask for image inpainting mode. If provided, aspect_ratio, width, and height inputs are ignored. seed** (integer, default: None): Random seed. Set for reproducible generation image** (string, default: None): Input image for image to image or inpainting mode. If provided, aspect_ratio, width, and height inputs are ignored. model** (string, default: dev): Which model to run inference with. The dev model performs best with around 28 inference steps but the schnell model only needs 4 steps. width** (integer, default: None): Width of generated image. Only works if aspect_ratio is set to custom. Will be rounded to nearest multiple of 16. Incompatible with fast generation height** (integer, default: None): Height of generated image. Only works if aspect_ratio is set to custom. Will be rounded to nearest multiple of 16. Incompatible with fast generation go_fast** (boolean, default: False): Run faster predictions with model optimized for speed (currently fp8 quantized); disable to run in original bf16 extra_lora** (string, default: None): Load LoRA weights. Supports Replicate models in the format <owner>/<username> or <owner>/<username>/<version>, HuggingFace URLs in the format huggingface.co/<owner>/<model-name>, CivitAI URLs in the format civitai.com/models/<id>[/<model-name>], or arbitrary .safetensors URLs from the Internet. For example, 'fofr/flux-pixar-cars' lora_scale** (number, default: 1): Determines how strongly the main LoRA should be applied. Sane results between 0 and 1 for base inference. For go_fast we apply a 1.5x multiplier to this value; we've generally seen good performance when scaling the base value by that amount. You may still need to experiment to find the best value for your particular lora. megapixels** (string, default: 1): Approximate number of megapixels for generated image How to Use Set up your Replicate API key in the workflow Configure the required parameters for your use case Run the workflow to generate other content Access the generated output from the final node API Reference Model: settyan/flash-v2.0.1-beta.10 API Endpoint: https://api.replicate.com/v1/predictions Requirements Replicate API key n8n instance Basic understanding of other generation parameters
by Yaron Been
Settyan Flash V2.0.0 Beta.10 AI Generator Description None Overview This n8n workflow integrates with the Replicate API to use the settyan/flash-v2.0.0-beta.10 model. This powerful AI model can generate high-quality other content based on your inputs. Features Easy integration with Replicate API Automated status checking and result retrieval Support for all model parameters Error handling and retry logic Clean output formatting Parameters Required Parameters prompt** (string): Prompt for generated image. If you include the trigger_word used in the training process you are more likely to activate the trained object, style, or concept in the resulting image. Optional Parameters mask** (string, default: None): Image mask for image inpainting mode. If provided, aspect_ratio, width, and height inputs are ignored. seed** (integer, default: None): Random seed. Set for reproducible generation image** (string, default: None): Input image for image to image or inpainting mode. If provided, aspect_ratio, width, and height inputs are ignored. model** (string, default: dev): Which model to run inference with. The dev model performs best with around 28 inference steps but the schnell model only needs 4 steps. width** (integer, default: None): Width of generated image. Only works if aspect_ratio is set to custom. Will be rounded to nearest multiple of 16. Incompatible with fast generation height** (integer, default: None): Height of generated image. Only works if aspect_ratio is set to custom. Will be rounded to nearest multiple of 16. Incompatible with fast generation go_fast** (boolean, default: False): Run faster predictions with model optimized for speed (currently fp8 quantized); disable to run in original bf16 extra_lora** (string, default: None): Load LoRA weights. Supports Replicate models in the format <owner>/<username> or <owner>/<username>/<version>, HuggingFace URLs in the format huggingface.co/<owner>/<model-name>, CivitAI URLs in the format civitai.com/models/<id>[/<model-name>], or arbitrary .safetensors URLs from the Internet. For example, 'fofr/flux-pixar-cars' lora_scale** (number, default: 1): Determines how strongly the main LoRA should be applied. Sane results between 0 and 1 for base inference. For go_fast we apply a 1.5x multiplier to this value; we've generally seen good performance when scaling the base value by that amount. You may still need to experiment to find the best value for your particular lora. megapixels** (string, default: 1): Approximate number of megapixels for generated image How to Use Set up your Replicate API key in the workflow Configure the required parameters for your use case Run the workflow to generate other content Access the generated output from the final node API Reference Model: settyan/flash-v2.0.0-beta.10 API Endpoint: https://api.replicate.com/v1/predictions Requirements Replicate API key n8n instance Basic understanding of other generation parameters
by Yaron Been
Barbacoaexpert1 Ai Haircuts AI Generator Description None Overview This n8n workflow integrates with the Replicate API to use the barbacoaexpert1/ai-haircuts model. This powerful AI model can generate high-quality other content based on your inputs. Features Easy integration with Replicate API Automated status checking and result retrieval Support for all model parameters Error handling and retry logic Clean output formatting Parameters Required Parameters prompt** (string): Prompt for generated image. If you include the trigger_word used in the training process you are more likely to activate the trained object, style, or concept in the resulting image. Optional Parameters mask** (string, default: None): Image mask for image inpainting mode. If provided, aspect_ratio, width, and height inputs are ignored. seed** (integer, default: None): Random seed. Set for reproducible generation image** (string, default: None): Input image for image to image or inpainting mode. If provided, aspect_ratio, width, and height inputs are ignored. model** (string, default: dev): Which model to run inference with. The dev model performs best with around 28 inference steps but the schnell model only needs 4 steps. width** (integer, default: None): Width of generated image. Only works if aspect_ratio is set to custom. Will be rounded to nearest multiple of 16. Incompatible with fast generation height** (integer, default: None): Height of generated image. Only works if aspect_ratio is set to custom. Will be rounded to nearest multiple of 16. Incompatible with fast generation go_fast** (boolean, default: False): Run faster predictions with model optimized for speed (currently fp8 quantized); disable to run in original bf16 extra_lora** (string, default: None): Load LoRA weights. Supports Replicate models in the format <owner>/<username> or <owner>/<username>/<version>, HuggingFace URLs in the format huggingface.co/<owner>/<model-name>, CivitAI URLs in the format civitai.com/models/<id>[/<model-name>], or arbitrary .safetensors URLs from the Internet. For example, 'fofr/flux-pixar-cars' lora_scale** (number, default: 1): Determines how strongly the main LoRA should be applied. Sane results between 0 and 1 for base inference. For go_fast we apply a 1.5x multiplier to this value; we've generally seen good performance when scaling the base value by that amount. You may still need to experiment to find the best value for your particular lora. megapixels** (string, default: 1): Approximate number of megapixels for generated image How to Use Set up your Replicate API key in the workflow Configure the required parameters for your use case Run the workflow to generate other content Access the generated output from the final node API Reference Model: barbacoaexpert1/ai-haircuts API Endpoint: https://api.replicate.com/v1/predictions Requirements Replicate API key n8n instance Basic understanding of other generation parameters
by Yaron Been
Monexia Nietgoed AI Generator Description None Overview This n8n workflow integrates with the Replicate API to use the monexia/nietgoed model. This powerful AI model can generate high-quality other content based on your inputs. Features Easy integration with Replicate API Automated status checking and result retrieval Support for all model parameters Error handling and retry logic Clean output formatting Parameters Required Parameters prompt** (string): Prompt for generated image. If you include the trigger_word used in the training process you are more likely to activate the trained object, style, or concept in the resulting image. Optional Parameters mask** (string, default: None): Image mask for image inpainting mode. If provided, aspect_ratio, width, and height inputs are ignored. seed** (integer, default: None): Random seed. Set for reproducible generation image** (string, default: None): Input image for image to image or inpainting mode. If provided, aspect_ratio, width, and height inputs are ignored. model** (string, default: dev): Which model to run inference with. The dev model performs best with around 28 inference steps but the schnell model only needs 4 steps. width** (integer, default: None): Width of generated image. Only works if aspect_ratio is set to custom. Will be rounded to nearest multiple of 16. Incompatible with fast generation height** (integer, default: None): Height of generated image. Only works if aspect_ratio is set to custom. Will be rounded to nearest multiple of 16. Incompatible with fast generation go_fast** (boolean, default: False): Run faster predictions with model optimized for speed (currently fp8 quantized); disable to run in original bf16 extra_lora** (string, default: None): Load LoRA weights. Supports Replicate models in the format <owner>/<username> or <owner>/<username>/<version>, HuggingFace URLs in the format huggingface.co/<owner>/<model-name>, CivitAI URLs in the format civitai.com/models/<id>[/<model-name>], or arbitrary .safetensors URLs from the Internet. For example, 'fofr/flux-pixar-cars' lora_scale** (number, default: 1): Determines how strongly the main LoRA should be applied. Sane results between 0 and 1 for base inference. For go_fast we apply a 1.5x multiplier to this value; we've generally seen good performance when scaling the base value by that amount. You may still need to experiment to find the best value for your particular lora. megapixels** (string, default: 1): Approximate number of megapixels for generated image How to Use Set up your Replicate API key in the workflow Configure the required parameters for your use case Run the workflow to generate other content Access the generated output from the final node API Reference Model: monexia/nietgoed API Endpoint: https://api.replicate.com/v1/predictions Requirements Replicate API key n8n instance Basic understanding of other generation parameters
by darrell_tw
Workflow Description This workflow demonstrates how to use the LINE Messaging API to handle two scenarios: Replying to a user's message using a reply token. Sending a push message to a specific LINE user using their user ID. Key Features Webhook Integration: Receives and processes incoming messages from LINE using a webhook. Conditional Logic: Checks if the received event type is a message and handles it accordingly. Reply Message: Automatically responds to the user's message using the LINE reply token. Push Message: Sends a test message to a specific LINE user using their unique user ID. Pre-Configuration To simplify the setup process, create a Header Auth credential in n8n: Name**: Authorization Value**: Bearer {line token} This will authenticate all API requests to the LINE Messaging API. Node Configurations 1.1. Webhook from LINE Message Purpose**: Captures incoming events from the LINE Messaging API. Configuration**: HTTP Method: POST Path: {n8n-webhook-page} 1.2. If Condition Purpose**: Checks if the received event type is message. Configuration**: Condition: {{ $json.body.events[0].type }} equals "message" 1.3. Line: Reply with Token Purpose**: Replies to the user's message using the LINE reply token. Configuration**: Method: POST URL: https://api.line.me/v2/bot/message/reply JSON Body: { "replyToken": "{{ $('Webhook from Line Message').item.json.body.events[0].replyToken }}", "messages": [ { "type": "text", "text": "收到您的訊息 : {{ $('Webhook from Line Message').item.json.body.events[0].message.text }}" } ] } 2.1. Manual Trigger: Test Workflow Purpose**: Triggers the workflow for testing the push message functionality. Configuration**: No additional setup required. 2.2. Edit Fields Purpose**: Prepares the unique LINE user ID for the push message. Configuration**: Field: line_uid: Uxxxxxxxxxxxx 2.3. Line: Push Message Purpose**: Sends a test message to a specific LINE user. Configuration**: Method: POST URL: https://api.line.me/v2/bot/message/push JSON Body: { "to": "{{ $json.line_uid }}", "messages": [ { "type": "text", "text": "推播測試" } ] } 工作流程描述 此工作流程展示如何使用 LINE Messaging API 處理兩種情境： 使用 reply token 回應使用者的訊息。 使用使用者的 user ID 發送 推播訊息。 主要功能 Webhook 整合：透過 Webhook 接收並處理來自 LINE 的訊息。 條件邏輯：檢查接收到的事件類型是否為訊息並進行處理。 回應訊息：使用 LINE 的 reply token 自動回覆使用者的訊息。 推播訊息：使用 LINE User ID 向指定用戶發送測試訊息。 預先設定 為簡化設定流程，請在 n8n 中建立 Header Auth 憑證： 名稱**：Authorization 值**：Bearer {line token} 此設定將用於認證所有 LINE Messaging API 的請求。 節點設定 1.1. Webhook from LINE Message 用途**：接收來自 LINE Messaging API 的事件。 設定**： HTTP 方法：POST 路徑：{n8n-webhook-page} 1.2. If 條件判斷 用途**：檢查接收到的事件類型是否為 message。 設定**： 條件： {{ $json.body.events[0].type }} 等於 "message" 1.3. Line: Reply with Token 用途**：使用 LINE reply token 回應使用者訊息。 設定**： 方法：POST URL：https://api.line.me/v2/bot/message/reply JSON 主體： { "replyToken": "{{ $('Webhook from Line Message').item.json.body.events[0].replyToken }}", "messages": [ { "type": "text", "text": "收到您的訊息 : {{ $('Webhook from Line Message').item.json.body.events[0].message.text }}" } ] } 2.1. 手動觸發：測試工作流程 用途**：測試推播訊息功能。 設定**：無需額外設定。 2.2. Edit Fields 用途**：準備推播訊息所需的 LINE 使用者 ID。 設定**： 欄位： line_uid：Uxxxxxxxxxxxx 2.3. Line: 推播訊息 用途**：向特定 LINE 使用者發送測試訊息。 設定**： 方法：POST URL：https://api.line.me/v2/bot/message/push JSON 主體： { "to": "{{ $json.line_uid }}", "messages": [ { "type": "text", "text": "推播測試" } ] } 完成示意圖 (Storyboard Example):
by Davide
1. How it Works This n8n workflow automates fine-tuning OpenAI models through these key steps: Manual Trigger**: Starts with the "When clicking ‘Test workflow’" event to initiate the process. Downloads a .jsonl file from Google Drive Upload to OpenAI**: Uploads the .jsonl file to OpenAI via the "Upload File" node (with purpose "fine-tune"). Create Fine-tuning Job**: Sends a POST request to the endpoint https://api.openai.com/v1/fine_tuning/jobs with: { "training_file": "{{ $json.id }}", "model": "gpt-4o-mini-2024-07-18" } OpenAI automatically starts training the model based on the provided file. Interaction with the Trained Model**: An "AI Agent" uses the custom model (e.g., ft:gpt-4o-mini-2024-07-18:n3w-italia::XXXX7B) to respond to chat messages. 2. Set up Steps To configure the workflow: Prepare the Training File: Create a .jsonl file following the specified syntax (e.g., travel assistant Q/A examples). Upload it to Google Drive and update the ID in the "Google Drive" node. Configure Credentials: Google Drive: Connect an account via OAuth2 (googleDriveOAuth2Api). OpenAI: Add your API key in the "OpenAI Chat Model" and "Upload File" nodes. Customize the Model: In the "OpenAI Chat Model" node, specify the name of your fine-tuned model (e.g., ft:gpt-4o-mini-...). Update the HTTP request body (Create Fine-tuning Job) if needed (e.g., a different base model). Start the Workflow: Use the manual trigger ("Test workflow") to begin the upload and training process. Test the model via the "Chat Trigger" (chat messages). Integrated Documentation: Follow the instructions in the Sticky Notes to: Properly format the .jsonl (Step 1). Monitor progress on OpenAI (Step 2, link: https://platform.openai.com/finetune/). Note: Ensure the .jsonl file adheres to OpenAI’s required structure and that credentials are valid.
by Harshil Agrawal
This workflow demonstrates the use of the Split In Batches node and the Wait node to avoid API rate limits. Customer Datastore node: The workflow fetches data from the Customer Datastore node. Based on your use case, replace it with a relevant node. Split In Batches node: This node splits the items into a single item. Based on the API limit, you can configure the Batch Size. HTTP Request node: This node makes API calls to a placeholder URL. If the Split In Batches node returns 5 items, the HTTP Request node will make 5 different API calls. Wait node: This node will pause the workflow for the time you specify. On resume, the Split In Batches node gets executed node, and the next batch is processed. Replace Me (NoOp node): This node is optional. If you want to continue your workflow and process the items, replace this node with the corresponding node(s).