class AnushkaSagar:
    name       = "Anushka Sagar"
    university = "Lovely Professional University, Punjab"
    degree     = "B.Tech — Computer Science & Engineering"
    cgpa       = 7.03

    focus      = ["AI Agents", "Federated Learning", "RAG Systems", "Backend APIs"]
    currently  = "Building privacy-preserving AI at the edge 🌐"
    published  = ["CSCT 2025 @ NIT Sikkim (Springer/Scopus)", "CGAIML2025 Book Chapter"]
    fun_fact   = "I make LLMs roast users for getting MCQs wrong 😈"

Privacy-preserving AI across distributed IoT edge devices — no raw data sharing.

• 🧠 Federated multi-agent framework trained on ~405K real-world IoT telemetry samples
• 📈 Improved accuracy 82% → 94.3% | Reduced latency 230ms → 150ms
• 🔐 Secured with HMAC-SHA256, JWT identity verification, Z-score & L2-norm anomaly detection
• Stack: Python Federated Learning JWT IoT Anomaly Detection

Modular agent pipeline that automates ad performance intelligence end-to-end.

• 🤖 4-agent system: Data → Insight → Evaluator → Creative
• 📉 Data drift detection to maintain model reliability over time
• 💡 LLM-driven analysis of ROAS, CTR, CPC validated by evaluation agent
• ⚡ Reduced manual analysis effort by ~70% via LangChain orchestration
• Stack: LangChain LLM Python Data Drift Multi-Agent

Deep learning on wearable sensor data — deployed with full DevOps pipeline.

• 🧬 Deep Neural Network achieving R² = 0.979, MAE = 0.803
• 🛠️ Infra as code with Terraform + monitoring via Nagios
• 🔄 Full CI/CD with Jenkins + Docker
• Stack: TensorFlow Puppet Terraform Jenkins Docker Nagios

Upload any PDF. Get brutally hard quiz questions. Get roasted if you fail.

• 📄 PDF → Chunk → Embed → FAISS vector store
• 🔍 Hybrid Search (vector similarity + keyword) to minimize hallucination
• 🦙 MCQ generation via LLaMA / DeepSeek through Ollama (fully offline)
• 😈 Savage UI that roasts the user at every step
• Stack: LangChain FAISS Ollama Streamlit nomic-embed-text PyMuPDF

Production-ready FastAPI backend with auth, ORM, migrations, and deployment.

• 🔐 OAuth2 + JWT authentication (HS256/RS256), password hashing
• 🗄️ SQLAlchemy ORM + PostgreSQL + Alembic migrations
• 🚀 Deployed on Ubuntu VM with Gunicorn + Nginx + Certbot (HTTPS)
• Stack: FastAPI SQLAlchemy PostgreSQL Alembic Nginx Gunicorn

GPT-2 product description generator refined through BLEU, ROUGE-L & human feedback.

• 🔁 Iterative feedback loop: auto scores + manual ratings → combined reward
• 📐 Combined = 0.7 × (Manual/10) + 0.3 × (BLEU + ROUGE-L)/2
• 🔭 Groundwork for RLHF fine-tuning pipeline
• Stack: GPT-2 Transformers NLTK ROUGE Matplotlib

• 🤖 Fundamentals of AI Agents Using RAG and LangChain — IBM | Coursera (Nov 2025)
• 🔬 Generative AI Engineering and Fine-Tuning Transformers — IBM | Coursera (Nov 2025)
• 🧪 IBM Machine Learning Specialization — IBM | Coursera (Oct 2024)
• ☁️ Cloud Computing with AWS — Gokboru Tech Pvt. Ltd (Jul 2024)
• 🧠 Generative AI with Large Language Models — AWS | Coursera (May 2024)

"Building AI that works at the edge, thinks in agents, and never hallucinates."