Five days into the Building Agentic AI Application Challenge with Decoding Data Science, I’ve taken AgroVn AI from an empty repo to a functional, domain-specific assistant. Here is what went into Days 1 through 5: Multi-Agent Core: Orchestrated Groq (Qwen-27B) with LangGraph to retrieve verified crop calendars and extension data. Lean & Fast: Achieved ~90% accuracy on verified queries with ~1.8s streaming latency at roughly $0.002 per query. Strict Guardrails: Enforced deterministic safety rules so the agent explicitly states "Not found" instead of guessing crop treatments. Interface Integration: Connected the backend pipeline to a mobile-friendly Gradio interface with custom table parsing. Heading into Days 6 to 8 to finalize security checks, optimize runtime, and deploy. What guardrails do you rely on when deploying AI in high-stakes domains? Decoding Data Science DDS Business Circle Mohammad Arshad #AgenticAI #AgroTech #Python #Langchain #LangGraph #RAG #MultiAgent #Gradio #LLMOps #MachineLearning

Successfully completed the day 2 of my agronomist agentic ai application !!

Watching an idea I started with gradually take shape through building, experimenting, making mistakes, and improving has been an incredible learning experience. I’m especially grateful to the Decoding Data Science (DDS) mentors and @Mohammad Ahmad for the thoughtful feedback and suggestions during Tuesday’s demo session this week. Their perspectives are helping me make Kindred better as I move toward the finish line. 🙏 Just a couple more days to go, and I’m genuinely excited to see what the final product becomes! ❤️ #Kindred #DecodingDataScience #AgenticAI #BuildInPublic #LearningByDoing #AIInnovation

Today was about taking ProDiag AI V2 from a working prototype toward a more complete maintenance product. The focus was on strengthening the Agentic Maintenance workflow and making sure the different components work together correctly. Today I worked on: ⚙️ Maintenance Agent → Maintenance Plan 🔧 AI-recommended Spare Parts integration 💰 Parts, labour & total maintenance cost calculation 👨‍🔧 Engineer-in-the-loop approval workflow 📋 Approved proposal → Work Order generation 🔗 Linking Work Orders with alerts and fault events 🐛 Debugging and fixing data-flow issues across the workflow One important improvement was making sure the recommended spare parts and their actual costs flow correctly through the maintenance proposal and into the final Work Order. For example, a replacement fan blade, labour cost, and total estimated maintenance cost can now be carried through the approval process instead of being lost between different stages. The AI doesn't create a Work Order by itself. AI recommends → Engineer reviews → Engineer approves → Work Order is created. That's an important step toward making ProDiag AI useful in a real industrial maintenance environment. With 2 days left, the focus now shifts toward final testing, documentation, presentation, and preparing the project for submission. 🚀 #BuildInPublic #AIAgents #AIChallenge #PredictiveMaintenance #IndustrialAI #AI #Maintenance #AgenticAI #Engineering #SmartManufacturing #Industry40 #BuildWithAI

Happy to share that - i completed day 01 successfully!