"Building the future of edge AI - one face at a time 🤖"
class LokeshPatra:
def __init__(self):
self.role = "Research Scholar + Associate DSML Developer"
self.focus = ["Edge AI", "Computer Vision", "IoT Systems", "Ethical AI"]
self.current_research = "Low-power facial recognition for resource-constrained environments"
self.education = "Data Science & Machine Learning"
def current_work(self):
return {
"project": "BioSentinel Edge",
"technologies": ["ESP32", "FastAPI", "React", "TensorFlow Lite", "GANs"],
"deadline": "December 3, 2025",
"innovation": "Synthetic GAN datasets for bias-free AI training"
}
def skills(self):
return {
"languages": ["Python", "C++", "JavaScript", "SQL"],
"ml_frameworks": ["TensorFlow", "PyTorch", "scikit-learn", "OpenCV"],
"backend": ["FastAPI", "Flask", "Node.js", "SQLAlchemy"],
"frontend": ["React", "TypeScript", "TailwindCSS"],
"iot": ["ESP32", "Arduino", "MQTT", "I2C/SPI protocols"],
"cloud": ["AWS", "Azure", "Docker", "PostgreSQL"],
"tools": ["Git", "PlatformIO", "Jupyter", "VS Code"]
}- Phase 4: Building React TypeScript dashboard for real-time face detection monitoring
- Phase 5: Integrating 100k+ synthetic faces from StyleGAN2 + FFHQ datasets
- Phase 6: Quantizing MobileFaceNet (FP32 → INT8) for ESP32 deployment
- Phase 7: Implementing Faiss vector similarity search for 1000+ identities
- Generative AI: StyleGAN2, LCA-GAN for synthetic face generation
- Model Optimization: TensorFlow Lite quantization, pruning, knowledge distillation
- Edge Computing: On-device inference with <2W power budget
- Ethical AI: Bias mitigation, fairness metrics, GDPR compliance
- System Design: Hybrid edge-cloud architectures for IoT
- 3/7 Phases Complete - BioSentinel Edge on track for Dec 3 deadline
- Real-time Face Recognition - 90% accuracy at 20 FPS on ESP32
- Ethical AI Pioneer - 100% synthetic dataset (zero real biometric data)
- Low-Power Design - 1.5W average consumption (battery-operable)
- End-to-End System - ESP32 → MQTT → FastAPI → Database verified
- Production Backend - 120+ identities, 320+ events logged in real-time
 Edge AI Low-power inference TensorFlow Lite Model quantization ESP32 deployment |
 Computer Vision Face detection Biometric analysis Multi-face tracking Real-time processing |
 Ethical AI Synthetic datasets GAN generation Bias mitigation Privacy-first design |
 IoT Systems Sensor integration MQTT protocols Hybrid architectures Resource optimization |
+ Generative AI (StyleGAN2, LCA-GAN) for privacy-preserving training
+ Edge Computing: Deep learning on microcontrollers (<1MB RAM)
+ Ethical Facial Recognition with 100% synthetic data
+ Hybrid Edge-Cloud Architectures for IoT scalability
+ Model Optimization: Quantization, Pruning, Knowledge Distillation
Hybrid edge-cloud facial recognition for resource-constrained environments
Tech: ESP32, FastAPI, React, TensorFlow Lite, Faiss, StyleGAN2
Status: Phase 3/7 Complete | Deadline: Dec 3, 2025
Highlights:
- <500ms end-to-end latency
- <2W power consumption
- 100% synthetic training data
- Multi-modal biometrics (face + HR + GSR)
-
I'm working on a 6-week research sprint to build a complete facial recognition system
-
I use 100% synthetic data because I believe in ethical AI
-
I optimize models to run on devices with <1MB RAM
-
My current project runs on <2 watts of power
-
I'm building technology for disaster response and remote environments
-
"I smile when I blink!" 😊
- Complete BioSentinel Edge (7 phases)
- Publish research paper on edge-based facial recognition
- Deploy production system for real-world testing
- Contribute to open-source computer vision projects
- Write technical blog series on ethical AI + edge computing
⭐ From lightxLK - Building the future of edge AI, one commit at a time!