# Projects & Engineering Portfolio — Nikhil Sai Pagidimarri

Short, concise overview of AI projects and models developed by Nikhil Sai Pagidimarri.

---

## 1. KareOS — Clinical Decision Support Platform
- **Organization:** Kraionyx AI ([kraionyx.com](https://kraionyx.com/products/kareos))
- **Tech Stack:** Multi-Agent AI Architecture, 20+ Medical AI Agents, Medical Imaging Analysis, HIPAA-Conscious Workspace
- **Highlights:**
  - Engineered clinical decision support suite powering doctor-facing AI workflows.
  - Deployed 20+ specialist medical AI agents for diagnosis assistance.
  - Integrated real-time medical image segmentation and ambient clinical documentation (Svaani).

---

## 2. Doclave — Professional Network for Healthcare
- **Organization:** Kraionyx AI ([kraionyx.com](https://kraionyx.com/products/doclave))
- **Highlights:**
  - Built professional networking platform for doctors and healthcare professionals.
  - Enabled medical knowledge exchange, peer collaboration, and verified professional profiles.

---

## 3. Svaani — Ambient Healthcare Voice Intelligence
- **Organization:** Kraionyx AI ([kraionyx.com](https://kraionyx.com/products/svaani))
- **Highlights:**
  - Developed ambient clinical voice AI converting doctor–patient conversations into structured notes in real time.
  - Reduced administrative documentation burden for clinicians.

---

## 4. Intelligent RAG Conversational AI Chatbot
- **Tech Stack:** Python, Pinecone, OpenAI text-embedding-3-large, FastAPI, MySQL, Hybrid Search
- **GitHub:** https://github.com/Nikhiluuuuuuuuu
- **Highlights:**
  - Architected a production RAG pipeline indexing 10,000+ documents with a MySQL ETL engine.
  - Delivered sub-second query latency with hybrid dense and keyword search.
  - Reached 95%+ accuracy.
  - Improved relevance scores by 40% and reduced hallucinations by 60%.

---

## 2. Brain Tumor Segmentation using ResUNet+
- **Tech Stack:** TensorFlow, Keras, ResUNet+, Attention Gates, GPU Infrastructure, Mixed Precision
- **Dataset:** BraTS20 MRI Dataset
- **Highlights:**
  - Built a multi-class brain tumor segmentation model with attention gates.
  - Achieved 98.17% accuracy and a 0.7570 Dice coefficient.
  - Optimized GPU mixed precision training over 20 epochs.
  - Real-time inference under 2 seconds per scan.

---

## 3. Hybrid MPC and PPO Autonomous Rocket Landing
- **Tech Stack:** JAX, Flax, Python, Model Predictive Control (MPC), PPO
- **Highlights:**
  - Combined Model Predictive Control with deep reinforcement learning.
  - Reached 94.7% landing success with 0.62m position accuracy.
  - Reduced training time by 62% over pure RL methods.
  - Sustained 78.3% success under ±20% mass variations and 80% thrust uncertainty.

---

## 4. Smart Diabetes Management System
- **Tech Stack:** Python, Scikit-learn, Logistic Regression, Naive Bayes, Decision Tree, Gradient Boosting
- **Organization:** NextHub Technologies Pvt Ltd
- **Highlights:**
  - Built an ensemble Combined Stacking Model for diabetes risk prediction.
  - Reached 92% prediction accuracy, surpassing baseline models by 8%.