Hi, I’m K V Jaya Harsha, a third-year undergraduate at IIIT Raichur with a strong interest in Applied Artificial Intelligence, Machine Learning, and R&D. I focus on building end-to-end ML systems — from data engineering and model development to deployment and monitoring — emphasizing scalability, modular design, and reproducibility.
Currently, I’m working as an AI Engineer Intern at Oscowl AI, where I build GenAI, RAG, and MLOps solutions, translating research ideas into production-ready AI systems. Previously, I served as PR Head for student initiatives, leading communications and outreach strategies.
| Programming Languages | Data Engineering / Big Data | Databases | Data Science / ML / Analytics |
|---|---|---|---|
     |
   |
     |
          |
| Web Frameworks | Cloud / Hosting / DevOps | Containers / Orchestration | CI/CD / Version Control |
|---|---|---|---|
  |
       |
  |
    |
| Other Tools / Libraries |
|---|
       |
A deep learning framework for predicting b-ion and y-ion fragment probabilities directly from peptide sequences using Graph Neural Networks (GNNs). Peptides are modeled as graphs where amino acids form nodes and peptide bonds define edges, enabling the model to capture both local and global structural dependencies.
Trained on the large-scale Pep2Prob benchmark dataset, the model outputs probability distributions across 78 fragment ion channels. An interactive Streamlit-based visualization platform allows real-time inspection of predicted fragmentation patterns.
A GenAI-powered Clinical Decision Support System designed to assist clinicians by transforming multimodal, unstructured clinical data into fact-grounded summaries and evidence-backed diagnostic insights.
The system acts as an intelligent co-pilot that:
- Reduces cognitive overload
- Improves clinical context awareness
- Provides explainable, evidence-linked outputs
The core novelty lies in a hybrid reasoning architecture combining:
- Deterministic semantic rules
- Strong Large Language Model (LLM) reasoning
- Multi-source Retrieval-Augmented Generation (RAG)
- Agentic tool orchestration
A robust hybrid segmentation framework for underwater imagery, integrating:
- DIP-based preprocessing for enhancement
- Multi-method segmentation strategies
- Vision Transformer (ViT)-driven patch refinement
The pipeline improves segmentation accuracy and robustness in challenging underwater environments with poor visibility and noise.
💡 Feel free to explore, clone, or contribute to any of these projects!
