MultiModal MultiMedia production space for creative souls, dancing through the beats of inspiration and vibrant visuals. Let your story be a tapestry of love, unity, and presence. Sol8, the AI-powered way to shine.

Building a platform with AI to make immigration easy and accessible to everyone :)

ML interview coaching, mock interviews for L5/6 ML engineer roles

Keywords: Multimodal AI, Self-supervised learning, Representation Learning, ActivityRecognition, Ray-ban Stories, Auto-capture feature, IMU2CLIP Designed, led, and developed multimodal AI solutions for AR/VR devices (context engine AI for smart glasses):* Tech lead: Cascading policies for multimodal context engine AI* Tech co-Lead: Autocapture interesting moments from smart glasses* Tech lead: Motion sensors (IMUs) efforts for context engine AI* Multimodal human activity recognition for AR/VR glasses* Representation learning using contrastive training (for IMU and Audiomodalities <-> Video, Text )* Music entity personalization and contextual ranking for entity resolutionPatents: * Active listening for assistant systems* Multi-device mediation for assistant systems* Dynamic Content Rendering Based on Context for AR and Assistant Systems* Task execution based on real-world text detection for assistant systems* Auto-Capture of Interesting Moments by Assistant SystemsPapers: * IMU2CLIP: https://arxiv.org/abs/2210.14395

•• Developed NLP and ASR solutions for Samsung's 1st voice assistant product (Bixby), available in over 200 countries •• Tech stack: Tesnorflow, pytorch, python, flask, spaCy, scikit-learn, AWS, S3 •• ML models: BiLSTM, CRF, Attention, Memory Networks, Transformer, Embeddings (BERT, Glove, Word2vec), CNN, Random Forest•• Projects: • The next-generation NLU parser for Bixby Product, to support multiple intents across multiple domains • Tech: dependency parsing, POS tagging, multi-task learning, transfer learning, random forest • Named Entity Recognition Service, to recognize and search named entities in real time ( for domains such as Music with millions of entities)  • Tech: NER, hybrid character and word embeddings, transfer learning • Text normalization to spoken format for benchmarking ASR results from different APIs (Bixby, Google Assistant, Alexa, Cortana) • On-device speech recognition system running on mobile phones (POC) • End-to-end deep learning speech recognition system for dictation (POC) • Noisy label training algorithms for intent classification (Product) • Task oriented dialogue breakdown detection system (Research)

Contributed to Samsung's Augmented Reality projects by developing machine learning and computer vision pipelines including: Head Pose Tracking, Dynamic Vision Sensors, and Sensor Fusion• Embedded deep learning development for real-time object recognition and tracking on mobile device for augmented reality applications • Localization and mapping on mobile device for 6DOF head pose tracking in VR/AR• Event based machine vision algorithm development for dynamic vision sensors

- Research Area: Applied Machine Learning, Siganl Processing , and Computer Vision for Bio-Robotic Applications

Designed and evaluated ML projects (supervised and unsupervised learning), with applications in computer visionTaught supplementary classes on classification and clustering Graded Exams and HWs

Wearable Controllers for Virtual Reality:- Developed software for integration of wearable motion controllers with IMU and EMG sensors (Nod, Myo) for hand gesture and motion input with Galaxy Gear VR (GVR) framework

- Conducted research in RGB-D mapping with mobile devices and depth sensors- Developed software for hand interaction with virtual reality (VR) using mobile device IMUs with Google cardboard - Conducted research in pattern recognition for food composition identification using spectroscopy