class ShubashisMete:
def __init__(self):
self.alias = "Suvo"
self.university = "SRM Institute of Science & Technology, Chennai"
self.degree = "B.Tech Computer Science & Engineering — Final Year"
self.cgpa = 9.36 / 10.0
self.location = "India 🇮🇳"
self.research_domains = [
"Federated Learning for Healthcare AI",
"Multimodal RAG Pipelines (LangGraph + BM25 Hybrid)",
"De Novo Drug Discovery — Graph Diffusion Models",
"Cross-Modal Contrastive Learning (TamilCLIP)",
]
self.engineering_stack = [
"MLOps Pipelines → Evidently AI + MLflow + CI/CD",
"Microservices → RabbitMQ + Redis + Stripe + Docker",
"LLM Fine-tuning → GRPO, QLoRA, Unsloth",
"Agent Systems → A2A Protocol + MCP + LangGraph",
]
self.open_to = [
"ML Engineer", "AI Agent Engineer",
"Research Intern", "Data Scientist",
]
self.north_star = "Research × Engineering = Civilizational Impact ⚡"
def current_focus(self):
return {
"🔨 building" : "A2A MCP Multi-Agent System + VoxPipeline",
"📄 publishing" : "Scopus-indexed ML research papers",
"🎯 applying" : ["Amazon ML Summer School", "Sarvam AI", "Ericsson"],
"☕ fun_fact" : "Trained a crop yield model at 3am 🌾",
}
 🤖 Building A2A MCP Multi-Agent System — LangGraph orchestration, autonomous tool-use & dynamic task decomposition |
 🧬 Researching De novo drug discovery for SARS-CoV-2 Mpro via graph diffusion models + active learning |
 📝 Publishing MedAI federated clinical ensemble — targeting Scopus-indexed journal (in review) |
 🏆 Competing SIH 2025 National Finalist · Amazon ML Summer School applicant |
🏥 MedAI — Federated Clinical Intelligence System
A privacy-preserving clinical decision support system powered by a Soft-Voting Ensemble (XGBoost + LightGBM + Random Forest) trained via Federated Learning across simulated multi-hospital nodes. Features MC Dropout Bayesian uncertainty, differential privacy (ε-δ guarantees), and physician-interpretable SHAP explanations — served via a FastAPI + React dashboard. Targeting Scopus-indexed publication.
Key Results: Multi-node federated training · Differential privacy · SHAP explainability · Clinical-grade uncertainty quantification
🌾 Kisan Yield Guard — Production MLOps Pipeline
Production-grade MLOps crop yield prediction system designed for real-world agricultural deployment. Integrates Evidently AI for automated data + model drift detection, MLflow for full experiment lineage, and GitHub Actions CI/CD with retraining triggers. Served via FastAPI with containerized deployment.
Key Results: Automated drift monitoring · Full experiment tracking · CI/CD retrain pipeline · Agricultural-domain ready
💊 De Novo Drug Discovery — SARS-CoV-2 Mpro
Generative molecular design pipeline targeting the SARS-CoV-2 main protease (Mpro) using graph diffusion models coupled with active learning for sample-efficient molecular optimization. Binding affinity is estimated via GNN-based molecular property prediction, validated on ESOL and BACE benchmarks.
Key Results: Graph diffusion generation · Active learning loop · GNN binding prediction · ESOL/BACE validation
🤖 A2A MCP Multi-Agent System
Domain-agnostic Agent-to-Agent coordination framework implementing the Model Context Protocol (MCP). Agents autonomously decompose tasks, select and invoke tools, and synchronize through a LangGraph finite-state-machine orchestrator backed by Groq/Llama ultra-low-latency inference.
Key Results: MCP-native architecture · Dynamic task decomposition · Autonomous tool-use · LangGraph FSM orchestration
🎙️ TamilCLIP — Cross-Modal Contrastive Learning
Cross-modal contrastive learning system aligning Tamil speech (Wav2Vec2 encoder) with Tamil text (IndicBERT) via InfoNCE loss — directly inspired by CLIP's vision-language alignment paradigm, now applied to low-resource Indian language audio-text understanding.
Key Results: Speech-text alignment · InfoNCE contrastive loss · Low-resource Indian language NLP · CLIP-inspired architecture
🛒 SkyBook — Domain-Agnostic Microservices Platform
6-service event-driven microservices booking platform with async inter-service communication via RabbitMQ, distributed session caching via Redis, real-time payment flows via Stripe, and full-stack deployment on Render + Vercel.
Key Results: 6 decoupled services · Async event-driven design · Redis caching · Stripe payments · Cloud-deployed
| 🏅 Achievement | 📋 Details | 📅 Year |
|---|---|---|
| 🥇 Smart India Hackathon Finalist | National Finalist — Top 40 teams across India | 2025 |
| 🎓 NPTEL · Machine Learning | Certified — IIT Kharagpur (Elite Track) | 2024 |
| 🎓 NPTEL · Database Management Systems | Certified — IIT Madras | 2024 |
| 📊 CGPA 9.36 / 10.0 | B.Tech CSE — SRM Institute of Science & Technology | Ongoing |
| 📝 Research Paper (In Review) | MedAI — Federated clinical ensemble · Scopus-indexed target | 2025 |
| 🔬 Research Paper (In Review) | Parasite detection via EfficientNet-B3 · IEEE target | 2025 |
╔══════════════════════════════════════════════════════════════════════════════╗
║ 🧠 Research Domains ║
╠═══════════════════════╦══════════════════════╦═════════════════════════════╣
║ 🏥 Healthcare AI ║ 🧬 Drug Discovery ║ 🤖 Agentic Systems ║
║ ───────────────── ║ ───────────────── ║ ───────────────── ║
║ Federated Learning ║ Graph Diffusion ║ A2A Coordination ║
║ MC Dropout Bayes ║ GNN Property Pred. ║ MCP Protocol ║
║ Differential Priv. ║ Active Learning ║ LangGraph FSM ║
║ Clinical Ensembles ║ Mpro Binding Pred. ║ Autonomous Tool Use ║
╠═══════════════════════╬══════════════════════╬═════════════════════════════╣
║ 🌾 MLOps ║ 🗣️ Multimodal NLP ║ 📈 Quant Finance ║
║ ───────────────── ║ ───────────────── ║ ───────────────── ║
║ Drift Detection ║ Cross-Modal CLIP ║ Hawkes Processes ║
║ MLflow Tracking ║ RAG Pipelines ║ QR-PPO RL Trading ║
║ CI/CD Retraining ║ IndicBERT + Wav2Vec ║ HMM Regime Detection ║
║ Feature Stores ║ BM25 Hybrid Rerank ║ Cryptocurrency Fraud ║
╚═══════════════════════╩══════════════════════╩═════════════════════════════╝
Open to research collaborations · internships · full-time ML/AI roles
"Design intelligent systems. Publish meaningful research. Engineer for civilizational scale."
⚡ Powered by curiosity, caffeine & 3am model training runs · suvomx1999 ⚡