Backend & ML student focused on security systems, real-time architectures, and applied NLP. Diploma graduate (CSE) → B.Tech CSE (AI & ML) at KMIT, Hyderabad (lateral entry). Open to internships.
| Project | Live | GitHub | Highlights |
|---|---|---|---|
| Prompt Injection Detection | Repo | 3-model ensemble · 95.26% recall · hybrid remediation · 99.84% utility · Zenodo preprint | |
| PhishShield | Demo | Repo | 18,684 samples · 97.19% offline · ~80 - 85% on 100 live emails · SHAP · Chrome MV3 |
| MusicFlow | Demo | Repo | 30+ REST APIs · Helmet.js · Socket.IO · Redis · collaborative filtering |
Research preprint: Zenodo — doi:10.5281/zenodo.20640595
Offline pipeline that detects prompt injections in email text and returns sanitized content for LLM agents to continue using — no external API calls.
Fine-tuned RoBERTa, SecureBERT, and DistilBERT on 5,995 samples. 3-model ensemble reaches 95.26% recall. Hybrid learned + regex remediation preserved 99.84% mean content retention on detector-flagged injected samples. Deployed via FastAPI.
Tech: Python · FastAPI · RoBERTa · SecureBERT · DistilBERT · HuggingFace Transformers · BIO tagging · Kaggle T4
Trained on 18,684 samples (TF-IDF + SecureBERT + MuRIL). Improved real-world detection from 42% → ~80–85% on 100 live emails (20+ edge cases fixed). FastAPI backend, fast explainability (SHAP optional), Chrome Extension (MV3), Docker CI.
Tech: Python · FastAPI · React · TypeScript · Docker · SecureBERT · MuRIL · SHAP · SQLite · Drizzle ORM · Zod · pnpm
30+ REST APIs with JWT, RBAC, Helmet.js security headers, and rate limiting across listener + admin React apps. Redis cache-aside with in-memory fallback. Socket.IO real-time sync. Collaborative filtering + AI playlists (MongoDB retrieval only — no hallucinated tracks).
Tech: React 19 · Node.js · Express · MongoDB · Redis · Socket.IO · JWT · Helmet.js · Cloudinary
Python · FastAPI · Node.js · Express · React · TypeScript · MongoDB · Redis · Socket.IO · Docker · JWT · HuggingFace Transformers · Scikit-learn · GitHub Actions