2025-09-14 –, Track 1
Machine learning is moving closer to the edge—but how close can you really get with Python and a ₹500 board? This session dives into running real-time ML inference on ultra-low-cost microcontrollers like the ESP32 using MicroPython and TinyML.
You’ll learn how to deploy models that classify audio, detect gestures, or monitor anomalies—all without an internet connection, operating system, or expensive tooling. This talk demystifies edge ML by showing how Python can still play a meaningful role in deeply constrained environments. If you're working in embedded systems, edge computing, or low-power AI, this session shows how to unlock a new layer of intelligence at the edge—one tiny inference at a time.
Why Edge ML Matters (5 min)
- ML use cases in constrained environments: health, energy, agriculture
- Why inference at the edge is essential (privacy, latency, connectivity)
TinyML and Python: What’s Possible? (5 min)
- Overview of TensorFlow Lite for Microcontrollers (TFLite Micro)
- Why Python still matters when deployment is in C
- Role of MicroPython and bridging model deployment to embedded
Building the Pipeline (10 min)
- Data collection and preprocessing using MicroPython
- Training models in standard Python using scikit-learn or TensorFlow
- Converting and quantizing models for microcontrollers
Deploying to a ₹500 Board (ESP32 / STM32) (5 min)
- Flashing the model to the device
- Using uTensor or C-compiled model blobs callable from MicroPython
- Triggering inference in real-time via GPIO/sensor inputs
Demos and Lessons Learned (5 min)
- Real-world use case: gesture classification / anomaly detection
- Limits, trade-offs, and tooling gaps in current Python-TinyML pipelines
- What to automate, what to hardcode
- Basic Python experience (basic scripting and file handling)
- High-level understanding of machine learning (training vs inference)
- Some familiarity with embedded boards (e.g., ESP32, Arduino) is a plus but not required
Beginner
Hi, my name is Pratik Kumar Panda. I am Currently working at RedHat as an SRE. Primarly working on Openshift Dedicated and ROSA.
In my free time, you can find me building drones and hobby robots - tinkering with Raspberry Pis, Arduinos, and automation with Python
I am currently working with Deloitte as a Software Engineer (HCSM - Infra and Cloud Management). I mainly deal with NOC and Cloud for my team.