AI/ML Researcher · Gen-AI · Deep Learning · Autonomous Systems
Pioneering the future of artificial intelligence through cutting-edge research and innovation. Published researcher bridging autonomous systems, computer vision, and intelligent healthcare.
"You have no responsibility to live up to what other people think you ought to accomplish. I have no responsibility to be like they expect me to be. It's their mistake, not my failing."
I'm Ashish Kishore, a Decision Scientist at Mu Sigma with a foundation in Electronics and Telecommunication Engineering. I specialise in building intelligent systems that bridge advanced AI research and real-world impact, with expertise spanning Machine Learning, Deep Learning, Computer Vision, Natural Language Processing, Generative Artificial Intelligence, Cloud Computing, Edge AI, IoT, and Embedded Systems.
During my academic career, I authored 3 papers in Q1 journals and presented at an international IEEE conference, spanning autonomous vehicles with state-of-the-art accuracy in Real-Time Traffic Sign Recognition, AyuSeva, a medium language model-powered AI medical chatbot recognised by healthcare professionals for its innovation in disease prediction, and Project ORION (Optimised Robotic Intelligence for Operative Neural-empathy), a privacy-preserving multimodal cognitive robot integrating deep facial expression recognition and speech sentiment intelligence with Edge AI for real-time empathetic human-robot interaction.
I've had the privilege of delivering keynote talks at Google Bangalore, sharing insights on AI in autonomous navigation, intelligent healthcare, and human-AI interaction.
"It is a profound and necessary truth that the deep things in science are not found because they are useful; they are found because it was possible to find them."
Co-author. Designed a novel real-time autonomous driving system using a custom CNN architecture from scratch with sensor fusion on Jetson Nano (45ms inference) and Arduino control. Achieved 99.68% on CTSD and 99.63% on GTSRB.
Read Paper ↗ GitHub Code ↗1st Author. Designed and implemented the complete D2B2C-IIFNN architecture from scratch, integrating DenseNet-style feature extraction, dual attention, BiLSTM, and Conv1D-based multimodal fusion. 99.52% symptom and 93.91% NLP accuracy with perfect Critical-tier recall. Approved by Amrutha Hospital.
Read Paper ↗ GitHub Code ↗1st Author. Created a privacy-preserving multimodal cognitive robot on-device. Combines secure HOG-SVM + ResNet-34 face verification (98.7% accuracy), a custom FER-EfficientNetV2 emotion classifier (95.33%), and localized Gemma-2 conversational module aligned via DoRA and DPO with CT-GraphRAG memory (~22 FPS on edge).
1st Author. YOLOv8x-based automated counting framework. Implemented custom Albumentations data augmentation and Streamlit deployment. Achieved 99.28% Precision, 85.71% Recall, 94.37% mAP@50, and 81.01% mAP@50-95 with 2.8ms single-user latency. Enables early yield estimation 2 weeks before harvest.
Read Paper ↗ View Presentation ↗I build at the intersection of intelligence and the physical world — systems that see, reason, adapt, and act in real environments.
My work spans autonomous perception, clinical decision intelligence, human-robot interaction, and radar target tracking. Each project begins with a problem worth solving and ends with results that can be measured, reproduced, and deployed.
Research, to me, is not a credential. It is a way of thinking.
Decision Scientist at Mu Sigma, one of the world's largest pure-play analytics and decision sciences companies. Joining to craft data-driven solutions that transform business decisions at scale.
AI Research Intern at LRDE-DRDO, Ministry of Defence, developing advanced deep learning based classification models and target manoeuvre tracking systems for strategic defence applications.
View certificate
Worked on cutting-edge space technology projects at U R Rao Satellite Centre, contributing to India's space research and satellite development programs.
View certificateBuilt models for SMS Spam Detection (LSTM), Customer Churn Prediction, Credit Card Fraud Detection, and Movie Genre Prediction using advanced NLP techniques.
View certificateDeveloped AI systems for Hand Gesture Recognition (99.97% accuracy), Calorie Estimation, House Price Prediction, and Customer Segmentation.
View certificate"Imagination is more important than knowledge. Knowledge is limited. Imagination encircles the world."
Delivered invited keynote on "AI in Autonomous Vehicles: Research to Real-World Impact" at Google India's Bangalore office, sharing insights with industry leaders.
Google Cloud Agentic AI Day, recognised for initiative and contribution in building the next generation of intelligent agents.
AyuSeva AI medical chatbot recognised for innovation in real-time disease prediction at national-level competition.
IEEE BIT CS Society recognition for high-performance ML model development, 1st place winner.
"The scientific man does not aim at an immediate result. He does not expect that his advanced ideas will be readily taken up. His work is like planting a seed for the future."
Problem: Human–robot interaction needs secure identity verification, reliable emotion understanding, and grounded long-term dialogue, while cloud-based personalization can expose sensitive user data.
Solution: An end-to-end on-device cognitive robot combining HOG-SVM+ResNet-34 face verification, FER-EfficientNetV2 emotion classifier, and localized Gemma-2 conversational pipeline with CT-GraphRAG memory retrieval entirely offline.
Problem: Existing clinical AI systems often handle structured symptoms and unstructured patient narratives separately, and many lack rigorous calibration, cost-sensitive safety checks, and reliable patient-facing dialogue.
Solution: AyuSeva, a multimodal clinical decision support system built around the D2B2C-IIFNN architecture for symptom-based disease prediction, paired with a confidence-gated generative conversational module for safe clinical guidance.
Problem: Autonomous vehicles need fast and reliable traffic sign recognition for safe driving, especially under changing lighting, occlusion, and road conditions.
Solution: A real-time CNN-based traffic sign recognition system integrated with an autonomous vehicle control platform.
Problem: Manual fruit counting and yield estimation in large orchards are labor-intensive, error-prone, and unreliable under occlusion, lighting variation, and dense foliage conditions.
Solution: YOLOv8x-based intelligent computer vision framework for automated jackfruit detection, counting, and yield estimation in smart agriculture.
Problem: Typing on physical keyboards limits interaction in AR/VR and accessibility contexts.
Solution: Real-time hand gesture recognition for mid-air typing system.
Problem: Single-factor authentication is vulnerable. Need multi-layer verification for high security.
Solution: Face ID + Fingerprint + Passcode system using multiple biometric modalities and Arduino hardware validation.
Delivered invited keynote presentation at Google India's Bangalore office, sharing insights on real-time traffic sign recognition, sensor fusion, and the future of autonomous navigation systems.
Presented research at the IEEE International Conference on Engineering Informatics 2025. Framework achieved 99.28% Precision, 85.71% Recall, 94.37% mAP@50, and 81.01% mAP@50–95.
Explore my verified skills and professional achievements, including badges and industry certifications, on my official Credly profile.
"Study hard what interests you the most in the most undisciplined, irreverent, and original manner possible."
Bengaluru, Karnataka · Graduated: May 2026
"We can only see a short distance ahead, but we can see plenty there that needs to be done."
"Innovation distinguishes between a leader and a follower." Steve Jobs
Download Resume ↓
