Yash Singh Yash-Singh607

⚡ YASH PRATAP SINGH ⚡

`AI & Data Science Student @ VIT Bhopal | Machine Learning & Computer Vision Developer`

🛸 Engineering Showcase: I am an AI & Data Science student at VIT Bhopal, specializing in Medical Computer Vision, Model Interpretability, and Temporal Natural Language Processing. Dedicated to translating machine learning research into robust, high-performance software systems.

🏆 Dynamic GitHub Achievements & Trophies

🚀 Core Focus & Profile Matrix

🔬 Core Focus & Engineering

🫁 Medical Diagnostics: Pulmonary X-Ray Segmentation & CLAHE contrast enhancement
🧠 Model Interpretability: Grad-CAM feature heatmaps for transparent diagnostic validation
💬 Temporal NLP: Trajectory, velocity, and acceleration tracking for sentiment drift
👁️ Edge Computer Vision: Real-time multi-stream video pipelines on NVIDIA Jetson

⚡ Quick Profile Matrix

🎓 Degree: B.Tech in Artificial Intelligence & Data Science
🏫 University: VIT Bhopal University
🏆 Activities: Hackathon Participant & LeetCode Solver
🤝 Status: Open for ML/CV Internship & Engineering Roles

🛠️ Technical Skill Matrix

💻 Core Languages

🧠 Machine Learning & Computer Vision

🌐 Frameworks & Tools

🧪 Featured Engineering Case Studies

🫁 01. PulmoScan — AI Pulmonary Diagnostic System

Medical Computer Vision & Visual Interpretability

🛑 Engineering Challenge: Radiograph datasets suffer from extreme class imbalance (rare pathologies vs. normal scans) and low local contrast that hides early lesions.
💡 Technical Solution & Impact: Applied CLAHE contrast normalization and Weighted Focal Loss with transfer learning architectures (ResNet/EfficientNet), achieving 94.2% ROC-AUC and reducing false-negative rates by 38%. Integrated Grad-CAM heatmaps for transparent visual verification.
🏷️ Tech Stack: PyTorch • OpenCV • Grad-CAM • Scikit-Learn • Python

🔍 View Deep Architecture Details

Preprocessing Pipeline: CLAHE (Clip Limit=2.0, TileGridSize=8x8) for X-ray contrast enhancement.
Loss Function: Focal Loss (alpha=0.25, gamma=2.0) to prioritize hard, ambiguous lesion samples over easy normal scans.
Interpretability Layer: Target layer activations hooked into Grad-CAM to produce 224x224 attribution heatmaps for clinicians.

🎭 02. Emotion Drift Detection System

Natural Language Processing & Sentiment Trajectory Dynamics

🛑 Engineering Challenge: Static sentiment classifiers analyze utterances independently in isolation, missing cumulative emotional decay in multi-turn customer conversations.
💡 Technical Solution & Impact: Built a temporal sliding-window transformer algorithm calculating sentiment velocity and acceleration over consecutive utterances using fine-tuned Hugging Face Transformer embeddings (89.6% F1-score), escalating high-risk chats 4.2 minutes earlier.
🏷️ Tech Stack: Python • Hugging Face Transformers • PyTorch • NLP • Pandas

🔍 View Temporal Math & Pipeline Details

Utterance Embeddings: Fine-tuned RoBERTa-base pooled output vectors.
Sliding Window Math: Window size w=3 tracking velocity vectors across message timestamps.
Escalation Trigger: Cosine distance threshold decay $> 0.45$ triggers priority queue routing.

👁️ 03. Shastra Eye — Real-Time AI Surveillance Pipeline

Edge Computer Vision & Real-Time Analytics

🛑 Engineering Challenge: Multi-stream high-fps video decoding on standard hardware creates CPU bottlenecks and frame drops during live threat detection.
💡 Technical Solution & Impact: Implemented CUDA multi-threaded frame buffers and TensorRT FP16 quantization to maintain 30+ FPS real-time performance with 40% lower latency on NVIDIA Jetson edge boards.
🏷️ Tech Stack: PyTorch • OpenCV • NVIDIA Jetson • Edge AI • Python

🧠 04. Visual Stress Detection System

Biomedical Signal Processing & Micro-Motion Analytics

🛑 Engineering Challenge: Intrusive contact sensors cause patient discomfort and artificial stress spikes.
💡 Technical Solution & Impact: Engineered a non-invasive vision pipeline utilizing dense Optical Flow at 60Hz to extract sub-visual facial micro-tremors, correlating 86% with standard HRV indicators.
🏷️ Tech Stack: Python • OpenCV • Optical Flow • Signal Processing • Scikit-Learn

🤖 05. Enterprise Knowledge Copilot

Intelligent Information Search & Document Retrieval

🛑 Engineering Challenge: Dense multi-page enterprise PDFs yield fragmented, out-of-context keyword search results.
💡 Technical Solution & Impact: Implemented vector similarity embeddings and FastAPI service architecture for rapid semantic context retrieval (320ms query latency).
🏷️ Tech Stack: LangChain • FastAPI • Python • Vector Search

📜 Certifications & Verified Credentials

Certification / Specialization	Issuing Organization	Credential Verification
🤖 Introduction to Machine Learning	Microsoft	[Verify Credential]
👁️ Fundamentals of Computer Vision	Vityarthi	[Verify Credential]
☁️ Cloud Computing	NPTEL (IIT Standard)	[Verify Credential]
☕ Java — Skill Certification	Roadmap.sh	[Verify Credential]
🗄️ SQL — Skill Certification	Roadmap.sh	[Verify Credential]
🧠 Applied Machine Learning in Python	Coursera / University of Michigan	[Verify Credential]
💼 ServiceNow Virtual Internship Program	ServiceNow	[Verify Credential]
📊 GenAI Powered Data Analytics Job Simulation	Forage	[Verify Credential]

📊 Live GitHub Analytics & Contribution Metrics

🤝 Connect & Collaborate

I’m open to collaborating on Medical Vision, Deep Learning, Edge AI, and Open-Source ML Projects!

⭐ Star the repositories if they helped you!

Navigation Menu