Rahul Kumar

M.Tech Student, Systems & Control Engineering
Indian Institute of Technology Bombay

email / github / linkedin

About

I am a Master of Technology student in the Systems & Control Engineering group at the Indian Institute of Technology Bombay (IIT Bombay). My work sits at the intersection of scientific machine learning, dynamical systems, and agentic artificial intelligence architectures.

Previously, I completed my B.Tech in Computer Science & Engineering from the School of Engineering and Technology at Nagaland University, where I graduated top of my class with a 9.62 CPI and was awarded the Institute Gold Medal for Academic Excellence.

Research Interests

Physics-Informed Neural Networks & Dynamical Systems Embedding governing ODEs/PDEs and conservation laws directly into neural network loss landscapes for data-free modeling and continuous-time system identification.

Machine Learning Theory & Kernel Methods Convex optimization in Multiple Kernel Learning (MKL), formulation of dual objectives with simplex constraints, and decision boundary analysis.

Agentic AI & Retrieval-Augmented Generation Multi-agent workflow orchestration (LangChain, Model Context Protocol), hybrid sparse-dense retrieval (BM25 + vector embeddings), and grounded reasoning.

Decentralized & Embedded Cyber-Physical Systems Edge sensor telemetry (ESP32, INA219), real-time WebSocket communication, and smart contract settlement protocols for distributed microgrids.

Publications & Preprints

Solana Powered Peer-to-Peer Energy Trading System Under Review

C. Y. Konyak, Rahul Kumar, M. Ozukum, R. Singh, and S. Hazarika

Proceedings of the 4th International Conference on Machine Intelligence for Research & Innovations (MAiTRI 2026), Springer Nature, May 2026.

Selected Projects

Physics-Informed Neural Networks for Dynamical Systems Spring 2025

M.Tech Seminar Project · Advised by Prof. Vivek Natarajan, Centre for Systems & Control, IIT Bombay

Formulated and implemented PINNs in PyTorch to solve non-linear dynamic equations without external training data by enforcing ODE physics residuals directly into the loss objective.
- Constructed composite loss combining physical ODE residuals and boundary condition penalties over 300 collocation points.
- Modeled spring-mass-damper dynamics via higher-order autograd, benchmarking against RK45 numerical baselines with an optimal L2 error of 2.30 × 10⁻⁴.
PyTorch · Scientific Machine Learning · ODE Systems · Autograd
EduRAG: Agentic Course Intelligence & QA Engine 2026

Independent Systems Research

Engineered an agentic RAG platform ingesting multi-format academic course materials (PDF, PPT, DOCX) with structured semantic indexing and modular query execution.
- Designed hybrid retrieval combining ChromaDB vector search with BM25 keyword matching and cross-encoder reranking.
- Orchestrated a multi-agent system coordinating Router, Retriever, Reasoning, Math Solver, and Citation agents in LangChain.
- Built an automated MCQ quiz generator with Ragas-based evaluation for faithfulness and context recall.
Python · FastAPI · Streamlit · LangChain · ChromaDB · BM25 · Docker
Solana-Powered Peer-to-Peer Energy Trading Platform 2024 – 2025

B.Tech Thesis · Advised by Dr. Chenlep Yakha Konyak, Dept. of CSE, Nagaland University

Architected a hybrid real-time microgrid energy trading platform linking physical edge sensor nodes to an on-chain automated settlement layer.
- Interfaced ESP32 microcontrollers with INA219 current sensors and relay switches, achieving >90% automated metering efficiency.
- Implemented sub-3-second transaction finality using Solana Devnet and SPL tokens linked to Phantom Wallet.
- Fabricated two physical hardware node prototypes demonstrating autonomous peer-to-peer energy dispatch upon cryptographic verification.
C++ · ESP32 · Solana · WebSockets · Hardware Prototyping · Python
Comparative Study of SVM Kernels & Multiple Kernel Learning Autumn 2025

Course Project: CS 725 (Foundations of Machine Learning) · Advised by Prof. Abir De, IIT Bombay

Investigated linear, polynomial, and RBF kernel formulations on high-dimensional multi-class datasets and implemented convex optimization for SimpleMKL.
- Benchmarked kernel decision boundaries on 581k-sample Forest Cover dataset using stratified sampling and 5-fold cross-validation.
- Formulated SimpleMKL dual optimization under simplex constraints, attaining 87.20% accuracy and outperforming single-kernel baselines.
Python · Scikit-learn · Convex Optimization · SimpleMKL
TechAura Official Web Platform & Event Portal 2023 – 2025

Institutional Web Engineering

Developed, deployed, and maintained the official festival web platform serving 300+ attendees with dynamic registration pipelines and server-side validation on Google Cloud Platform.

Python · Flask · JavaScript · GCP App Engine

Education

Indian Institute of Technology Bombay 2025 – 2027

Master of Technology in Systems & Control Engineering

Relevant Coursework: Foundations of Machine Learning (CS 725), Organization of Web Information (CS 728), Deep Learning: Foundations & Applications (CS 6012), Generative & Agentic AI (IE 624).
School of Engineering & Technology, Nagaland University 2021 – 2025

Bachelor of Technology in Computer Science & Engineering · CPI: 9.62 / 10.0
Don Bosco Higher Secondary School, Kohima 2019 – 2021

Higher Secondary School Leaving Certificate (Class XII), Science · NBSE · 84.60%
Bayavü Hill High School, Kohima 2017 – 2019

High School Leaving Certificate (Class X) · NBSE · 88.83%

Technical Skills & Certifications

Languages

C, C++, Python, SQL, JavaScript, Bash

Frameworks

PyTorch, Scikit-learn, FastAPI, Flask, React, LangChain

AI & Systems

Physics-Informed ML (PINNs), LLM Agents, Model Context Protocol (MCP), RAG, ChromaDB

Tools & Infra

Git, Linux, Docker, Google Cloud Platform (GCP), AWS EC2

Anthropic Certs

Model Context Protocol, Subagents, Agent Skills, Claude Code, AI Fluency