Loading Experience
PuneethAI/ML Researcher
Deep Learning Research Intern at IIT Kharagpur · B.Tech Data Science at MVGR · Building intelligent systems that bridge AI and social impact
I build
intelligent
systems.
From offline-first rural healthcare AI to real-time vibration signal classifiers, I build production-ready systems that solve real problems.
Bridging
AI &
social impact.
I believe technology should serve humanity. My projects focus on healthcare accessibility and educational empowerment.
Ready to see
my work?
Featured
Projects
Real-world AI systems I’ve built — from offline healthcare diagnostics to phishing detection and safety tech.
Rural Med AI
Offline-first Progressive Web App for rural ASHA workers — AI-powered skin condition classification (HAM10000 / TensorFlow.js) and wound segmentation (U-Net / ONNX), with emergency SMS alerts, multi-language support, and IndexedDB sync queue. Runs 100% offline after first load.
MedMitra
Comprehensive healthcare management system connecting doctors, health workers, and patients. Features AI-powered donor matching, voice-to-text prescriptions, entity extraction, QR health IDs, emergency SOS, and Cure Map visualization with role-based dashboards.
Browser Shield AI
Chrome extension with 9 rule-based detection methods and a 128-feature neural network for real-time phishing detection. Hybrid scoring (40% rules + 60% ML) achieves 95%+ accuracy — all processing is 100% local with zero data leaving the browser.
Raksha Kavach
Multi-modal personal safety system combining a wearable hardware ring trigger with an AI voice guardian. Uses multi-layered audio analysis for universal distress signals (screams, shouts) with failsafe redundancy and silent SOS alerts for real-world emergencies.
EduHub
Educational platform designed to enhance STEM learning through interactive content delivery, resource management, and collaborative tools for students and educators.
Where I've
worked
Deep Learning Research Intern
Department of Mining Engineering, IIT Kharagpur
Supervisor: Prof. Subhendu Mishra
Project: Deep Learning-Based Classification of Hydrocyclone Discharge States Using Vibration Signals
- Designed DCAF-Net (Dual-Domain Cross-Attention Fusion Network) — a novel deep learning architecture fusing time-domain 1D ResNet and frequency-domain SE-CNN branches via an 8-head cross-attention mechanism for real-time vibration-based hydrocyclone discharge state classification.
- Engineered 11 amplitude-preserving auxiliary features (log-RMS energy, spectral centroid, band energy ratios, Shannon entropy) to restore discriminative physical information removed by Z-score normalisation.
- Achieved 98.2% mean ternary classification accuracy under Leave-One-Configuration-Out (LOCO) cross-validation across 4 unseen hydrocyclone geometries — a rigorous industrial generalisation benchmark.
- Deployed a production-ready monitoring system: FastAPI inference server + WebSocket streaming + real-time browser dashboard with Signal Quality Index (SQI) gate and K-NN OOD detector; CPU-only, <100 ms latency.
- Confirmed statistical significance via McNemar's test (p < 0.001) across 2,963 evaluation windows, outperforming SE-Swin, TCN-BiLSTM, EfficientNet-1D, and Contrastive WaveNet baselines.
- Preparing manuscript for Elsevier submission: Engineering Applications of AI (IF 10.23) and Advanced Powder Technology (IF 4.52).
Turning ideas into
reality
Engineering student at MVGR College of Engineering pursuing B.Tech in Data Science with a strong foundation in Python, PyTorch, and Deep Learning. Recently completed a research internship at IIT Kharagpur designing novel neural architectures for industrial signal classification.
Proven track record of leadership as a National-level Chess player and Team Lead, with demonstrated expertise in developing intelligent systems and educational technology solutions. Passionate about leveraging AI for social impact.
Let's create
something amazing
Have a project in mind? I'd love to hear about it. Send me a message and let's start the conversation.
Or reach out directly