Jyoti Jyoti

I’m an AI & Research Engineer with hands-on experience in computer vision, multi-modal systems, deepfake detection, and LLM-integrated pipelines. I build end-to-end ML solutions that turn model outputs into actionable, real-world decisions.

At Kryoniq, I created benchmarking audio deepfake detection datasets and ran forensic feature analysis on real and deepfake audios. I designed and trained deep learning models achieving 96% prediction accuracy in synthetic speech detection across multi-language datasets, and I integrated LLM-based report generation to produce structured forensic analysis reports.

At APEPDCL, I’m building a CV pipeline for vegetation encroachment detection to enable early fault detection in power grid systems. I’m implementing depth estimation using MiDaS to quantify spatial distance between power lines and encroaching vegetation, and I’m integrating model outputs into a GIS-based infrastructure platform for real-time alerting and spatial asset management.

Before that, I delivered backend services for real-time KPI analytics and automated anomaly detection at Zoro Innovations, and I supported AI/ML & Data Science work at K.A.M.A.L.A. My projects span VoiceGuard (96% multilingual deepfake detection with CNNs), FloodSense (IoT + LiDAR flood monitoring with 92% accuracy), and TwinDrive (digital twin and predictive maintenance), alongside NLP work like MarianMT domain adaptation—work I’ve validated through awards and competitive recognition.