EXPERIENCE SYSTEMS PORTFOLIO SKILLS & ARCHITECTURE EDUCATION & HONORS CERTIFICATIONS GET IN TOUCH
PRODUCTION BENCHMARKS
5.06M+ ord/s NexusEngine P99 <500ns
•
93.8% Grounded Enterprise MCP (JSON-RPC 2.0)
•
9.85 Gbps AetherPlane DPDK/eBPF
•
59.7K+ txn/s BillDesk Kafka CDC <150ms
•
1.8ms NPU NeuraType DirectML INT8
•
$19.7M Saved Walmart Demand Analytics
•
0.9247 ROC-AUC PaySim GAT Mule Detection

Architecting High-Throughput Distributed & AI Systems

Distributed Systems Model Context Protocol (MCP) Low-Latency C++ & Java DAU, Gujarat (2027)

Mudit R is a Distributed Systems & AI Systems Engineer building institutional Model Context Protocol (MCP) gateways, sub-microsecond matching cores, kernel-bypass network planes, and resilient full-stack AI platforms. Based in India, focused on high-throughput data engineering and production-grade intelligence.

5.06M+ Orders / Sec
8.02 DAU CPI / 10
600+ DSA Solved
93.8% RAG Groundedness
MUDIT R // CORE Systems Ready

MUDIT R

Distributed Systems & AI Architecture

Java 22 • C++20 • Python Kafka CDC • MCP • PyG
Engineering opportunities?

Get in touch
Enterprise MCP Platform
Enterprise MCP Platform Preview
MCP JSON-RPC 2.0 • 93.8% Grounded • AST SQL
#1 Enterprise MCP Platform
#2 NexusEngine Matching Core
#3 Graph Fraud GNN Engine
#4 NeuraType Edge-AI
#5 NextChapter RAG Reader
#6 Walmart Demand Forecaster
#7 AetherPlane Data Plane
#8 IoT Telematics & Feast
PRODUCTION EXPERIENCE & RESEARCH
FINTECH DISTRIBUTED BACKEND & STREAMING

Software Engineering Intern

BillDesk • Mumbai, Maharashtra

Jul 2026 – Aug 2026
  • Architected a real-time NoSQL to RDBMS replication service in Java 17 & Spring Boot across Kafka topics, sustaining 59,700+ transactions/sec peak throughput with partitioned concurrency and sub-150ms P99 latency.
  • Engineered transactional multi-table fan-out for parent-child schemas using Spring TransactionTemplate, ensuring atomic relational commits and automated rollback on child write failures.
  • Implemented dynamic YAML schema mappings, array expansion, and tuned HikariCP connection pooling, eliminating connection starvation across 250,000+ benchmark transactions (cutting manual migration effort by 85%).
  • Containerized the multi-service test stack (Kafka, MySQL, MongoDB) via Docker Compose and established automated CI/CD pipelines, slashing setup time by 70%.
Java 17 Spring Boot Apache Kafka (CDC) TransactionTemplate HikariCP MySQL MongoDB Docker Compose
FINTECH RISK & QUANT ML RESEARCH

Summer Research Intern

Dhirubhai Ambani University (DAU) • Gandhinagar

May 2026 – Jun 2026
  • Researched alternative credit scoring for informal sector workers, modeling cash-flow volatility across high-frequency UPI streams across 1,200+ merchant cohorts.
  • Engineered an automated ML underwriting microservices pipeline in FastAPI with sub-50ms inference latency, evaluating Debt-Service Coverage Ratios (DSCR) to reduce default risk by 22%.
  • Integrated a Web Speech API voice ledger prototype enabling informal merchants to record voice-based micro-transactions with automated text normalisation and validation.
FastAPI Python UPI Stream Analytics Risk Scoring Cash-Flow Volatility Web Speech API
SYSTEMS PORTFOLIO (RANKED BY TECHNICAL COMPLEXITY)
#1 MODEL CONTEXT PROTOCOL (MCP) & SELF-HEALING RAG
GitHub ↗
Enterprise MCP Platform Preview P99 <15ms • 93.8% Grounded

Enterprise GenAI & Governed Model Context Protocol (MCP) Platform

Architected an Anthropic-compliant Model Context Protocol (MCP) server (JSON-RPC 2.0) with RBAC, orchestrating an autonomous ReAct agent across warehouse schema introspection, DTI metrics, and policy retrieval. Engineered a two-stage self-healing RAG pipeline combining dense vectors, BM25 (RRF), and neural Cross-Encoder re-ranking with automated hallucination repair achieving 93.8% groundedness and AST SQL injection defense.

Model Context Protocol JSON-RPC 2.0 FastAPI ReAct Agents Cross-Encoder BM25 + RRF AST SQL Parser
#2 ZERO-ALLOCATION MATCHING CORE (SUB-500NS)
GitHub ↗
NexusEngine Matching Core Architecture 5.06M+ ord/s • Sub-500ns P99

NexusEngine — High-Throughput Limit Order Book & Matching Core

Engineered an institutional Limit Order Book and Matching Core in Java 22 & C++20 sustaining 5.06M+ orders/sec with deterministic sub-500ns P99 latency. Implemented zero-allocation memory patterns using 64-byte cache-padded SPSC lock-free ring buffers, pre-allocated object pools, intrusive doubly-linked list price structures, and memory-mapped binary Write-Ahead Log (WAL).

Java 22 C++20 Lock-Free SPSC Cache Padded FIX 4.2 Gateway Binary WAL Zero Allocation
#3 GRAPH NEURAL NETWORKS & 6.36M NODES (FINTECH RISK & AML)
Live Demo ↗ GitHub ↗
Graph Fraud Detection Network Preview 0.9247 ROC-AUC • 3.28M Accounts • 75.0% P@100

Graph-Based Fraud & Mule Account Detection Engine

Constructed a directed topological transaction graph across 6.36M payment records (3.28M bank accounts, 2.77M directed edges) to isolate money-muling rings and asymmetric routing loops. Engineered 22 graph-structural and temporal features (PageRank centrality, K-core decomposition, 24h velocity spikes, cycle detection). Trained and benchmarked GCN, GraphSAGE, and GAT against gradient-boosted trees; engineered a hybrid GAT + XGBoost stacking ensemble with Focal Loss and CARE-GNN edge filtering countering 130:1 extreme class imbalance, achieving 0.9247 ROC-AUC and 75.0% Precision@100 with a sub-50ms FastAPI & Redis inference service.

PyTorch Geometric Graph Attention Networks (GAT) XGBoost Stacking CARE-GNN Filtering PageRank Centrality FastAPI Redis Cache CUDA C++
#4 ON-DEVICE DIRECTML NPU & INT8 QUANTIZATION
GitHub ↗
NeuraType Edge AI Preview 1.8ms NPU • 8.13 MB INT8

NeuraType — Edge-AI & On-Device Neural NLP Engine

Built an air-gapped on-device typing engine with zero network egress, preventing sensitive credential and text exfiltration. Quantized a neural language model into an 8.13 MB INT8 ONNX graph accelerated via DirectML (NPU/GPU), delivering sub-2ms local inference across rolling 20-word context windows with Win32 global hooks and real-time regex PII redaction.

PyTorch ONNX Runtime DirectML (NPU) INT8 Quantization Win32 Hook Air-Gapped Zero Egress
#5 PRODUCTION RAG & 1,000+ SESSIONS
Live App ↗ GitHub ↗
NextChapter Platform Preview 1,000+ Users • <50ms Navigation

NextChapter — AI-Powered Digital Reading & Context-Aware Assistant

Full-stack digital reading platform in React 18 & FastAPI with a virtualized PDF.js canvas engine and LRU caching, sustaining sub-50ms page navigation across 1,000+ active user sessions. Engineered an in-book grounded RAG assistant via Pinecone and Llama 3.3, enforcing dynamic page-cutoff guardrails to prevent spoilers with Server-Sent Events (SSE) token streaming.

React 18 FastAPI Llama 3.3 Pinecone RAG SSE Streaming Supabase RLS PDF.js
#6 QUANTITATIVE ECONOMETRICS & 42.8M RECORDS
GitHub ↗
Walmart Demand Forecasting Analytics Preview 42.8M Sales • $19.7M Saved

Hierarchical Demand Forecasting & Price Elasticity Platform

Built an enterprise quantitative forecasting engine across 42.8M sales records (30,490 SKUs across 10 supercenters), benchmarking LightGBM, SARIMA, Prophet, and Chronos-2 transformer models. Modeled econometric price elasticity (-1.45 discretionary vs -0.85 staples) and safety stock compression, quantifying a 25.7% inventory reduction and $19.7M USD in holding savings.

LightGBM Chronos-2 SARIMA / Prophet Price Elasticity FastAPI Streamlit
#7 C++20 KERNEL-BYPASS & 9.85 GBPS
Live Demo ↗ GitHub ↗
AetherPlane C++20 DPDK Data Plane Preview 9.85 Gbps • 0.42 μs Core

AetherPlane — Ultra-Low-Latency Network Data Plane & AI QoS Engine

Architected a multi-gigabit zero-copy kernel-bypass data plane in C++20 using DPDK & eBPF/XDP, achieving 9.85 Gbps line-rate forwarding and 0.42 μs core latency. Implemented an AI flow classifier executing sub-microsecond inference driving FQ-CoDel Active Queue Management to eliminate wireless bufferbloat, with 16-8 Radix LPM routing.

C++20 DPDK eBPF / XDP FQ-CoDel AQM Radix LPM Linux Kernel
#8 GCP CLOUD STREAMING & MLOPS
Live Demo ↗ GitHub ↗
IoT Telematics Cloud Architecture Preview 10,000+ msg/s • Sub-3ms Feast

Real-Time Cloud Telematics & Predictive Feature Store

Cloud telematics engine ingesting 10,000+ msg/s sensor streams from 30+ fleet assets via GCP Cloud Pub/Sub into a BigQuery Lakehouse. Deployed a dual-tier Feast feature store with offline BigQuery lakehouse and low-latency cache, delivering sub-3ms feature lookups at zero idle infrastructure cost with Terraform IaC automation on GCP Cloud Run.

GCP Pub/Sub BigQuery Feast Feature Store Terraform IaC Cloud Run FastAPI
TECHNICAL EXPERTISE & ARCHITECTURE

Engineering Tooling Matrix

Languages & Systems Core
Python C++ (20) Java (22/17) SQL (PostgreSQL, MySQL) TypeScript / JavaScript C
AI, LLM Systems & Model Context Protocol (MCP)
Model Context Protocol (MCP) Autonomous ReAct Agents Hybrid RAG (Dense + BM25 RRF) Neural Cross-Encoders AST SQL Security Guardrails PyTorch Geometric (GNN/GAT) ONNX Runtime DirectML (NPU) Pinecone Optuna
Distributed Backend & Data Streaming
Apache Kafka (CDC) Spring Boot FastAPI Zero-Allocation Memory SPSC Lock-Free Buffers Redis HikariCP MongoDB
Cloud Infrastructure & DevOps
Google Cloud Platform (GCP) BigQuery Lakehouse Cloud Run & Pub/Sub Feast Feature Store Docker Kubernetes Terraform (IaC) GitHub Actions CI/CD

Education & Honors

2023 – 2027

B.Tech in Information and Communication Technology

Dhirubhai Ambani University (DAU), formerly DA-IICT • Gandhinagar, Gujarat

CPI: 8.02 / 10.0
600+ DSA Problems solved across LeetCode & Codeforces; Codeforces Pupil rating.
Convener, The Debating Society of DAU (2025–2026): Adjudicated national parliamentary debates and mentored 80+ university debaters.
Qualcomm & IBM Certified: AI Model-to-App Development & Machine Learning with Python.
Systems Architecture: Contributed to high-throughput CDC data streaming, sub-microsecond matching cores, and edge NPU accelerators.
LICENSES & VERIFIED CREDENTIALS
Qualcomm Issued Aug 2026
Show credential

AI Upskilling Certificate: Hands-On Development from Model to App

End-to-end artificial intelligence and machine learning engineering covering deep neural network development, edge model optimization, and full-stack AI application deployment.

Artificial Intelligence (AI) Edge AI Model Optimization Deep Learning App Deployment
IBM Issued Aug 2026 • ID: 3561837769e245d6af7f3035472a370f
Show credential

Machine Learning with Python

Comprehensive certification in classical ML and statistical learning covering supervised & unsupervised algorithms, regression, classification, clustering, Scikit-Learn pipelines, and model evaluation.

Machine Learning Python Scikit-Learn Supervised Learning Statistical Modeling
COLLABORATION & CONTACT PORTAL

Let's engineer something extraordinary.

Available for Software Engineering, Distributed Systems, High-Throughput Infrastructure, and Production AI/ML opportunities.

mudit14127@gmail.com (click to copy) github.com/Mudit-R linkedin.com/in/Mudit-R
AI & MCP PLATFORM
Architecture Preview

Architecture Overview

Verified Benchmarks & SLA

Core System Flow

GitHub Repository ↗ Live Deployment ↗
Mudit-OS v3.5 [Systems Shell • Type 'help']
Mudit-OS v3.5 High-Performance Systems Shell Type 'help' to see all available commands. mudit@portfolio:~$ help Available commands: • bio : Engineer biography & background • experience : Inspect BillDesk CDC & DAU FinTech internships • projects : List all 8 systems in ranked order • mcp : Inspect Enterprise Model Context Protocol platform • nexus : View NexusEngine 5.06M+ orders/sec matching specs • neuratype : View NeuraType DirectML INT8 edge typing engine • walmart : View 42.8M sales hierarchical demand forecasting • metrics : View real-time production SLAs and benchmarks • skills : Matrix of languages, MCP, data & cloud tooling • certifications : View Qualcomm & IBM verified credentials • theme : Toggle between Obsidian Cyber Dark & Editorial Warm Pastel • contact : Display direct contacts (email, GitHub, LinkedIn) • clear : Clear terminal screen
mudit@portfolio:~$