AI / ML Engineer

Vanshaj Rawat

Building intelligent systems at the intersection of computer vision, deep learning, and real-time inference. Turning research into deployable AI.

View Projects Get In Touch
87ms DeepShield Latency
100% RL Agent Success Rate
2 Deployed Projects
ICAMC 2026 Conference
Technical Arsenal

Built to ship AI.

🧠
AI / ML Frameworks
PyTorch Scikit-learn OpenCV MediaPipe LibROSA NumPy Pandas Matplotlib
⚡
Generative AI & LLMs
Hugging Face Transformers MarianMT LangChain RAG Google Gemini
🔬
Domains
Computer Vision Deep Learning Reinforcement Learning NLP Deepfake Detection Multimodal AI
🛠️
Languages & Tools
Python C++ JavaScript TorchScript Gymnasium Streamlit Git
Work

Projects that run in prod.

Computer Vision ICAMC 2026
DeepShield
Python · OpenCV · MediaPipe · PyTorch · LibROSA · NumPy
  • Real-time deepfake detection analyzing lip-sync, pose dynamics, and phonetic-visual consistency simultaneously.
  • Three-module fusion pipeline: MediaPipe facial landmarks + LibROSA audio features + custom PyTorch fusion model.
  • Optimized with ThreadPoolExecutor, frame downsampling, and TorchScript — achieving 87ms latency vs 210ms (MesoNet) and 380ms (Xception).
↗ Live Demo
87ms inference latency
Reinforcement Learning
Taxi Route Optimization
Python · Gymnasium · Q-Learning · NumPy · Matplotlib
  • Tabular Q-Learning agent solving Gymnasium Taxi-v3 from scratch; 100% success rate and avg reward of +7.6 across 100 unseen episodes.
  • 500-state Q-table with ε-greedy exploration, Bellman equation updates, α=0.1, γ=0.99 — converges in under 5,000 episodes (~9 seconds).
  • Interactive dark-mode dashboard with live state explorer, episode replay, Q-value heatmaps, and learning curve visualizations.
↗ Live Demo
+7.6 avg reward
Machine Learning
Multiple Disease Prediction
Python · Scikit-learn · Streamlit · SVM · Logistic Regression · Pickle
  • Multi-model clinical web app integrating three independent ML models — diabetes (SVM), heart disease (LR), and Parkinson's (SVM) — in a single Streamlit interface.
  • Serialized trained Scikit-learn models with Pickle; real-time predictions across multiple clinical features with zero page reloads.
↗ GitHub
3 disease models
Background

Education & Certifications

B.Tech — Artificial Intelligence & Data Science
Galgotias University, Greater Noida
OCT 2022 — PRESENT
Intermediate (PCM)
Ursuline Convent Sr. Sec. School, Greater Noida
2021 — 2022
☁️
Generative AI Virtual Internship (10 weeks)
AICTE & Google Cloud · EduSkills
🏆
Hackastra Finalist
NIT Durgapur National Hackathon
🤖
IBM AI Fundamentals
IBM SkillsBuild
☕
Python Essentials 1, 2 · Cisco
Cisco Networking Academy
Let's Connect

Open to AI/ML roles.

Recent AI & Data Science graduate looking to build production-grade ML systems. If you're working on something challenging, let's talk.

✉️ vanshajxrawat@gmail.com 💼 linkedin.com/in/vanshaj-rawat 🐙 github.com/Vanshajrawat 📱 +91 93104 76956 🐙 leetcode.com/u/vanshajrawat/