AWS IoT Core demo with Lerobot SO-101 Arms using Micropython

Excited to share my latest project, where I applied Machine Learning to predict faults in industrial IoT networks. Technologies Used: Python, Random Forest, Pandas, CI/CD pipeline, GitHub Goal: Detect anomalies in IoT devices to improve reliability and prevent downtime Outcome: Built a predictive model capable of identifying potential faults across multiple IoT nodes This project helped me strengthen skills in network automation, AI, and practical ML deployment, bridging the gap between theoretical knowledge and real-world applications. Check it out on GitHub: https://lnkd.in/eGsRajze #MachineLearning #IoT #Python #NetworkAutomation #CI_CD #IndustrialIoT #DataScience #AI

In the world of real-time pipelines, we see more and more customers combining CrateDB with Apache Flink. This is a trend I am analyzing in this new blog post: Powering Real-Time Streaming Analytics with CrateDB and Apache Flink 🔗 https://lnkd.in/e3bGqQuJ I explain how CrateDB and Flink together help organizations that can’t wait for insights (whether from IoT devices, SaaS usage data, or operational metrics) to act in milliseconds instead of minutes or hours. Here’s what I cover: - Why traditional databases and batch pipelines can’t keep up at industrial scale ✅ How Flink manages high-velocity streaming and real-time event processing ✅ The power of CrateDB as a scalable SQL database supporting both streaming and historical analytics in one place ✅ Use cases where this stack drives real business value (from manufacturing to SaaS monitoring) 💬 I’d love to hear from you: ✅ Have you used Flink with SQL databases for real-time analytics in production? ✅ What challenges did you face around latency, scale, or architectural complexity? ✅ Do you see opportunities for GenAI or ML models in this kind of pipeline (real-time plus historical data)? Let’s discuss, share your thoughts below 👇 Thomas Gérard Chris Horsnell #RealTimeData #StreamingAnalytics #ApacheFlink #CrateDB #DataArchitecture #IoT #SaaS #OpenSource (Ververica | Original creators of Apache Flink®

More and more teams are pairing Apache Flink with CrateDB to power streaming pipelines that deliver insights in milliseconds. If you’re working with Flink, make sure you’re getting the most out of it — CrateDB is built for this. Check out this new post on how CrateDB + Flink help organizations move from reactive to real-time: 🔗 Powering Real-Time Streaming Analytics with CrateDB and Apache Flink. #RealTimeData #Flink #CrateDB #StreamingAnalytics #IoT #DataEngineering

Just launched: Edge AI Inference Platform Built on AWS Wavelength + IoT Greengrass, optimized for sub-10 ms latency at the 5G edge. We talk a lot about AI at scale this one’s about AI at the edge. Closer to the user. Faster than the cloud. Think: drones, factories, hospitals places where one millisecond matters. This project started as an experiment, ended up a full production-grade stack: ✅ Auto-scaling and cost-optimized ✅ Secure, monitored, fault-tolerant ✅ TensorRT-optimized models running on 5G Wavelength nodes It’s open-source and MIT-licensed. For anyone building real-time inference or federated learning at the edge, this might save you a few weekends. 👉 GitHub: https://lnkd.in/dbm5p5-D Sometimes innovation isn’t about “more compute.” It’s about less distance. #AWS #Serverless #EdgeAI #IoT #Greengrass #Wavelength #DevOps #OpenSource #CloudArchitecture

💡 NDJSON - The Unsung Hero of Real-Time Data Pipelines If you have ever built a data ingestion or IoT pipeline, you know JSON is easy to read… until you start streaming gigabytes of it. That's where NDJSON (Newline Delimited JSON) quietly changes the game. Each JSON object sits on its own line - no giant array, no memory overhead, no parsing nightmare. Just clean, streamable data. ⚙️ Why It Matters ✅ Perfect for streaming data ingestion (Kafka, Flink, Spark, Logstash, etc.) ✅ Enables line-by-line parsing - great for tailing logs or API responses ✅ Ideal for IoT telemetry, ETL pipelines, or microservice event logs ✅ Easy to debug - every event is valid JSON, independent of the rest It's the difference between processing 1 GB of data at once and processing millions of micro-events efficiently. 📊 Real-World Example In one ingestion pipeline, we switched from bulk JSON to NDJSON for sensor data: Parsing latency dropped ~40 % Stream buffer size reduced ~60 %\ Error recovery became instantaneous - no broken arrays, no retries. 🧠 Architect's Takeaway As systems scale, simplicity wins. NDJSON is not flashy - but it's battle-tested for distributed environments where every millisecond and every newline matters. 💬 Have you used NDJSON in your streaming or IoT projects? What's your favorite tool or use case for it? #SystemDesign #DataEngineering #IoT #Streaming #Architecture #Kafka #Flink #ETL #SagarThakkar

Predictive Maintenance using IoT & Machine Learning. I recently built an end-to-end predictive maintenance solution that predicts machine failures before they happen, helping reduce downtime and improve efficiency. What I built: IoT Sensors - Data ingestion & transformation ML Model - Trained to predict failures MLflow - Tracking experiments & metrics Prediction Pipeline - Serves real-time predictions Deployment - Flask app running on AWS EC2 (Dockerized) Results: Achieved 100% accuracy on test data Confusion Matrix shows perfect predictions for all cases #MachineLearning #AI #IoT #MLOps #DataScience #PredictiveMaintenance #AWS #Docker #MLflow

So new client, now challenges - unfortunately because of NDA restrictions, I have had to censor the #ErlangRed flow - but trust me, it works! Yes this is Erlang code, yes it uses Statemachines, yes it uses MQTT connectivity and it's visual - well behind the black censoring there are actually nodes! And the best bit? The flow (without the black bits) can just be copied and pasted into a Node-RED instance (that the clients knows and loves) and be run there - no change needed, no glue, no BS. The client uses a **lot** of #Erlang code and they asked me whether Erlang-Red could be of help to them. They already know #NodeRED and came across Erlang-Red (as you do) and what does the flow actually do? It provides a simulation for IoT devices that communicate with the base Erlang system. This is a proof of concept but after three days of work, I can basically replace a bunch of Python code with Erlang-Red flows. #ErlangRed #NodeRED #Erlang #IIoT #FunTimes

🌐 Building a Minimum Spanning Tree in Distributed Systems I recently studied how to find a Minimum Spanning Tree (MST) efficiently in a distributed environment — where data or nodes are spread across multiple machines. Using Borůvka’s Algorithm, each node selects its smallest outgoing edge, and components merge iteratively until one MST remains. This approach scales well for large graphs and distributed data networks. ⚙️ 💡 Real-world ML Insight: In sensor networks or distributed feature selection, MSTs help minimize redundancy and optimize communication paths — ensuring efficient data flow and model updates. It’s fascinating how classical graph algorithms still drive modern ML and IoT applications! 🚀 #GraphAlgorithms #MST #DistributedSystems #MachineLearning #DataEngineering #Python

Dynamic Load Balancing Across IoT Clusters (A Python Prototype) While debugging a distributed IoT deployment recently, I discovered two clusters were overloaded while others were almost idle. The imbalance sparked an IoT Cluster Balancer solution using Python, which; - Continuously analyzes cluster load (processing, throughput, or task count) - Detects imbalance across regions - Suggests minimal cost rebalancing, keeping the network efficient with minimal data movement Example scenario: C1: 70 | C2: 40 | C3: 90 | C4: 30 Total = 230 | Average ~ 57.5 Rebalance strategy: - Transfer 15 from C3 → C4 - Transfer 10 from C1 → C2 Result C1: 60 | C2: 50 | C3: 75 | C4: 45 Perfect balance isn't the goal, efficient equilibrium is. In real-world IoT networks, stability matters more than symmetry. Explore the source code on GitHub. #IoT #EdgeComputing #DistributedSystems #LoadBalancing #Python #SystemDesign #Innovation

Testing InfluxDB 3.0 in FactoryTalk Optix! First sensations using this new database version who brings new functionalities like SQL query language in a time stamp database. Even a Grafana web window is inserted inside FT Optix to display statistical process analytics. Risoul Es #rockwellautomation InfluxData Grafana Labs Docker, Inc #influxdb #grafana #docker #factorytalkoptix #ftoptix #iot #iiot #digitaltransformation #industry40 #edgecomputing #ai You can find the code here https://lnkd.in/dBgjWxsi https://lnkd.in/d3uFNtj4