Author: Shabbir Poonawala

  • Voice-to-JSON: Structuring Unstructured Audio.

    When it comes to transcribing audio recordings, manual listening is a slow and laborious process. I’ve developed a solution using OpenAI Whisper to quickly transcribe audio and an LLM to extract key fields like Name, Budget, and Urgency into a clean JSON object. This data is then pushed instantly to our CRM before the phone even hangs up.

    This process has significantly reduced the time spent on data entry and has improved the accuracy of our lead data. By automating this process, we’ve been able to make data-driven decisions and optimize our sales strategy. If you’re looking to streamline your sales process and improve your data accuracy, consider implementing a similar solution. Learn more about how we can help you achieve your sales goals.

  • The ‘Busy Tone’ is a Software Bug: Fix it with Exotel + n8n

    A physical phone line can only handle one call. A SIP trunk can handle thousands. I’ll show how I connect Exotel to n8n to spin up a new AI instance for every single incoming call, ensuring zero latency and zero waiting time.

    The ‘busy tone’ is not just a physical limitation, but a software bug that can be fixed. By leveraging Exotel and n8n, you can create an infinite capacity system that never says ‘busy’. Learn more about how ShahiRaj can help you solve this software bug.

  • Building a Quranic Vector Store: RAG for Scripture

    As a developer, I’ve always been fascinated by the challenges of searching for concepts in the Quran. Standard SQL search just doesn’t cut it, especially when it comes to understanding the semantic intent behind a user’s query.

    That’s why I decided to build a Quranic Vector Store using OpenAI Embeddings and a Vector Database. This allows the Digital Hafiz bot to understand the context and sentiment behind the user’s query, mapping their emotion to the relevant divine guidance.

    Here’s a high-level overview of the architecture:

    • Indexed the Quran using OpenAI Embeddings
    • Stored the embeddings in a Vector Database
    • Used a neural network to map the user’s query to the relevant embeddings
    • Retrieved the most relevant Ayats from the Vector Database

    By using this approach, the Digital Hafiz bot can provide a more personalized and meaningful experience for users, helping them to deepen their understanding of Islam and live a more meaningful life.

    Try the Digital Hafiz today and experience the power of Vector Search for yourself!

    #ShahiRaj #DigitalHafiz #QuranSearch #VectorSearch

  • Linking the Misri Calendar to n8n Triggers: A Technical Deep Dive

    As a developer, I’m always on the lookout for innovative solutions to real-world problems. The AlQuranAlKareembot’s ‘Miqaat’ automation feature sparked my interest, and I decided to dig deeper into its technical aspects.

    One of the key challenges I faced was linking the Misri Calendar to n8n triggers. The Gregorian scheduler simply doesn’t work for Hijri dates, making it essential to create a custom solution.

    I developed a ‘Miqaat Node’ – a custom Javascript function that checks the current date against the Misri Calendar algorithm. If the day matches Friday Eve, it triggers the specific Surah Kahf workflow.

    If you’re interested in exploring more technical aspects of the AlQuranAlKareembot, I encourage you to join the conversation and share your thoughts!

  • Regex for Finance: Parsing Text Commands in n8n

    I’ve been working on a project to automate expense tracking using n8n, a workflow automation tool. One of the key features is parsing text commands from Telegram messages.

    The text command might look like this: ‘Paid 2000 for server hosting’. My goal is to extract the amount, category, and narrative from this text and push it to Firestore/Sheets.

    I’ll break down the JavaScript Regex logic I use to achieve this. If you’re interested in learning more about regex and how to apply it to finance, stay tuned for the next part of this series.

    Check out the ShahiRaj blog for more technical articles on finance and workflow automation.

    #ShahiRaj #n8n #Regex

  • Integrating Uptime Kuma with Telegram via n8n

    I’m excited to share with you a simple yet powerful setup that will change the way you handle server monitoring. Say goodbye to email alerts and hello to high-priority notifications on Telegram!

    In this post, I’ll walk you through integrating Uptime Kuma webhooks with n8n, filtering out false positives, and sending critical alerts to a private Telegram channel. The best part? It’s a 3-node setup, making it easy to implement and maintain.

    Don’t rely on email alerts that you might miss. With this setup, you’ll receive timely notifications and stay on top of your server’s performance. Follow along and learn how to set up your own server monitoring system using n8n and Uptime Kuma. Get in touch if you have any questions or need help implementing this setup.

  • Webhooks vs. Polling: The Architecture of Real-Time Tracking.

    Most tracking apps lag by hours. I’ll show you how I use n8n to listen for Shipmozo Webhooks instantly. The moment the courier scans the packet, my database updates, and a WhatsApp message triggers. Latency: <2 seconds. Check out the code and architecture.

    #ShahiRaj #n8n #Shipmozo

  • Beyond Keywords: Semantic Search for Visitor Queries.

    Visitors to our website rarely use the right keywords. Instead, they ask vague questions like ‘Do you do that thing with the phones?’ That’s where semantic search comes in. With the help of Pinecone and OpenAI Embeddings, our bot can understand the intent behind the messy query and retrieve the correct service page from our vector store.

    But how does it work? Well, first we use Pinecone to index our vector database. Then, when a visitor asks a question, we use OpenAI Embeddings to generate a vector representation of their query. Finally, we use the Pinecone index to find the closest match in our vector database.

    This approach allows our bot to provide accurate and relevant answers to our visitors’ questions, even when they don’t use the right keywords.

    Learn more about our semantic search technology and how it can help you improve your customer experience.

    #ShahiRaj #SemanticSearch #AI

  • RAG for Objection Handling: Coding a Bot That Can Negotiate.

    Standard bots fail when a user says ‘It’s too expensive.’ But not ours. With Vector Search, we retrieve specific ‘ROI data’ and ‘Competitor Comparisons’ from a PDF knowledge base, allowing the LLM to counter objections with hard facts instantly.

    In this technical deep dive, I’ll show you how to code a bot that can negotiate and seal the deal. By combining RAG, Vector Search, and LLMs, we’ve created a bot that can handle objections and close deals.

    Want to learn more? Check out our ShahiRaj blog for the latest on RAG-powered chatbots and objection handling.

  • OCR is Dead. Long Live GPT-4o Vision.

    Old OCR tools failed on complex layouts, but I’ve found a game-changer in GPT-4o Vision. This AI-powered vision model inside n8n allows my bot to ‘look’ at an image (like a conference agenda) and output clean JSON meeting details, handling even blurry text perfectly.

    I’ll demonstrate how I integrated GPT-4o Vision into my bot’s workflow, showing you how it can extract meeting details from images and sync them with your calendar.

    Want to learn more about this cutting-edge technology? Check out our blog for the latest updates and tutorials.