Introducing Lumen AI — an AI-powered Dubai Utility Bill & Energy Assistant, built during the DDS Enterprise AI Bootcamp by Decoding Data Science. The problem it solves: Understanding a DEWA bill isn't easy. Electricity slabs, water charges, fuel surcharge, housing fee, sewerage fee, and VAT can make it confusing—especially for new residents. Lumen AI explains bills in plain language and provides accurate estimates using official DEWA tariff information. You can ask it: 🔹 "Why did my DEWA bill increase this month?" 🔹 "Estimate my bill for 3,500 kWh." 🔹 "How do electricity slab rates work?" 🔹 "How can I reduce my monthly energy bill?" Tech Stack ⚙️ LlamaIndex (RAG)📦 Pinecone🤖 OpenAI GPT-4o-mini 🌐 @Flask 🐳 Docker, Inc 🤗 Hugging Face Spaces What I learned ✅ Better data beats bigger models. ✅ RAG quality depends on the knowledge base. ✅ Building and deploying a real AI product is the best way to learn. 🚀 Live Demo: https://huggingface.co/spaces/nipunkavindaAI/LumenAI 💻 GitHub: https://github.com/NipunKavinda95/LumenAI Thanks to Mohammad Arshad and Decoding Data Science for another hands-on learning experience. DDS Business Circle Dubai Electricity & Water Authority - DEWA #AI #GenerativeAI #RAG #LlamaIndex #OpenAI #Pinecone #HuggingFace #DEWA #Dubai #BuildWithAI #DecodingDataScience #LearningInPublic

Pleased to share my dss chatbot which I made during July bootcamp https://03848a2eb9773d4f94.gradio.live/

Over the past two days, I’ve been diving deep into AI product development and Retrieval-Augmented Generation (RAG)—gaining both strategic frameworks and hands-on experience building real-world AI solutions. 🚀 Workshop 1: AI Product Thinking & RAG Foundations The opening session focused on the bridge between business requirements and technical execution—understanding how successful AI products are actually designed and built. Key Learnings: Idea Translation: Converting raw concepts into clearly defined AI project scopes. Problem-First Mindset: Identifying core business challenges before picking the tech stack. Data Strategy: Defining data requirements and effective ingestion strategies. LLM Evaluation: Comparing model capabilities, latency, and cost structures across different providers to select the best fit. Parameter Tuning: Mastering OpenAI configuration parameters like Temperature, Top-P, and Max Tokens. RAG Architecture: Understanding the end-to-end workflow 🤖 Workshop 2: Building a RAG Chatbot with LlamaIndex The second workshop was all about hands-on implementation in Google Colab, building a functional RAG pipeline from scratch. Tech Stack: ⚙️ LlamaIndex: Document ingestion, chunking, indexing, retrieval orchestration, and context management. ⚙️ Gradio: Rapid deployment of a clean, interactive chatbot user interface. Key Takeaway: LlamaIndex simplifies complex RAG engineering tasks that would otherwise require extensive custom boilerplate. It allows developers to focus on solving core business problems rather than managing underlying infrastructure. 💡 Practical Project: DDS HR Chatbot To apply these concepts directly, I built an HR Chatbot for DDS using the RAG architecture. Key Features & Impact: Retrieves answers directly from internal HR documents. Delivers accurate, context-aware responses. Minimizes hallucinations by grounding answers strictly in company knowledge. Demonstrates how domain-specific AI can transform internal knowledge access and employee support.

I had the privilege of attending a recent session with DDS AI and learned so much about thinking about your product when working with AI, Agentic AI vs RAG, Prompt Engineering, and using the appropriate language model for your use case. One major insight was that when building with AI you want to start with a real problem you are trying to solve instead of picking the latest and greatest model or trying to make another chatbot. We discussed how Agentic AI is capable of reasoning, using tools, having a memory, and acting on a plan. While RAG allows your AI to connect with verified documents to create more grounded answers. I’m also learning the hard way how important it is to just try things out, be your own user when testing, write down what works, and iterate. Just because one model performed better for one task doesn’t mean they will for another task. Select your model based on what the user needs and what your project requires. I’m an Electrical Engineering student, and these are lessons I will take with me as I start to brainstorm for my senior design project. I’ve been working on an idea for a while now. It’s an AI called CampusMate, an intelligent university assistant to help students keep track of classes, search academic documents, and plan their semester. Thanks to Decoding Data Science and the DDS Business Circle, the mentors that lead the sessions, and the AI community as a whole. I can’t wait to learn more, try new things, build, reflect on my successes and failures, and share my journey. #DDSNextGen #DDSAmbassador #DecodingDataScience #DDSBusinessCircle #AgenticAI #ArtificialIntelligence

Today I worked on improving Nexa AI’s backend and preparing the system for the next stage. What I did: ✅ Improved the career recommendation engine ✅ Cleaned the agent workflow ✅ Prepared Gemini AI integration ✅ Tested the system and fixed issues Building a strong AI application starts with creating a reliable foundation before adding more advanced features. 🚀 #AgenticAI #AIProjects