Aryan Gupta

FLAGSHIP ARCHITECTURES

Engineered for breakthrough performance.

Four core initiatives blending high-concurrency simulation, deep vision, and scalable cloud engineering.

Reinforcement Learning PyTorch · CUDA · Pymunk

MARL Tag: 3v3 Pursuit-Evasion

30 Million Timesteps. Relentless Self-Play.

95% Seeker win rate across 100 benchmark episodes

8x Subprocess environments parallelized on CUDA

Engineered a self-play multi-agent RL framework utilizing Proximal Policy Optimization (PPO) with Generalized Advantage Estimation (GAE). Designed a 102-dimensional raycast spatial awareness perception network and custom physics-reward engine.

Launch interactive arena

National Research Ministry of Electronics & IT

Covert Steganography & Dual-Value QR

Invisible to Adversaries. Crisp to Authenticators.

2-Tier Distance-based cryptographically signed payload

CV ControlNet + Stable Diffusion generative blending

Pioneered aesthetic full-image error-correction QR codes for covert data transmission and anti-phishing verification during government research fellowship at MeitY. Enabled multi-factor defensive threat mitigation.

Inspect research specs

Distributed Systems MADDPG · A* Pathfinding

Vehicle Fleet Optimization Engine

Optimal Trajectories. Minus the Congestion.

-20% Passenger wait time reduction at peak load

100+ Concurrent autonomous agents coordinated

Built an autonomous taxi dispatch simulation combining dynamic A* heuristic routing with Multi-Agent Deep Deterministic Policy Gradients (MADDPG) for non-stationary collaborative transit environments.

Explore routing logic

Cloud Infrastructure IBM Cloud · ARIMA · Flask

Scalable Time-Series Forecasting

Predictive Intelligence. Live on IBM Cloud.

Cloud End-to-end containerized model hosting

ARIMA Robust predictive stationarity pipeline

Cleaned, validated, and normalized volatile enterprise time-series datasets. Trained stationary ARIMA forecasting algorithms and deployed auto-scaling REST endpoints on IBM Cloud infrastructure.

Review deployment

INTERACTIVE EXPERIMENTAL BENCHMARK

Autonomous 3v3 Arena. Live in your browser.

A direct real-time simulation of Aryan's multi-agent pursuit-evasion architecture. Watch autonomous seekers hunt evaders with raycast spatial awareness—or take manual control.

3v3 LIVE SIMULATION

ALGORITHM PPO + GAE (Self-Play)

SEEKER WIN RATE 95.4%

ACTIVE RAYS 102 Dimensions

EVADERS ACTIVE 3 / 3

SEEKERS ARMED 3 / 3

SIM TICK 0

WASD or ←↑→↓ to evade hunters in Manual Mode · Click canvas to plant acoustic obstacle

Perception Matrix

102 raycast sensors per agent providing distance, relative angular velocity, and obstacle occlusion vectors.

Distributed Workers

Parallelized actor-critic updates over 8 distinct subprocess environments accelerated via CUDA PyTorch tensor pipelines.

Curriculum Reward Shaping

Non-linear distance-squared penalty curve balanced against evasion stealth bonuses to prevent policy degeneration.

UNDER THE HOOD

Architectural stack. Built for resilience.

From hardware acceleration and CUDA kernels to cloud-native microservices and defensive security.

Reinforcement Learning & AI

Autonomous Decision Architectures

PPO + GAE
MADDPG Multi-Agent
Self-Play Dynamics
PyTorch
scikit-learn
Keras
Reward Engineering
Pymunk Physics

Core Optimizer: AdamW (lr=3e-4)

GAE Lambda / Gamma: 0.95 / 0.99

Benchmark Verification: 95% Win Rate @ 100 Episodes

Computer Vision & GenAI

Aesthetic & Defensive Synthesis

Stable Diffusion
ControlNet QR Conditioning
Dual-Payload Steganography
OpenCV
Error-Correction Tuning
Distance Cryptography

Payload Encoding: Cryptographically Signed Dual-Layer

Research Agency: MeitY, Government of India

Use-Case Focus: Anti-phishing & Covert Channels

Systems, Cloud & Data

High Throughput & Concurrency

Python (Proficient)
CUDA Acceleration
C / C++
JavaScript / WebGL
IBM Cloud
Weights & Biases (W&B)
Flask & REST APIs
MySQL & Unix/Linux

OS & Shell: Unix / Linux, Bash, PowerShell

Version Control & CI: Git, GitHub Actions, Docker

Data Pipelines: NumPy, Pandas, Time-Series ARIMA

Leadership & Execution

Team Direction & Mentorship

ACM MUJ Projects Head
CyberSpace Research Member
Hackathon Team Director
Technical Mentorship
Sprint Planning

Team Built & Led: 15+ Engineers (40+ Interviewed)

Project Delivery: 90% Completion Success Rate

Competition Record: Multiple Hackathon Finalist Spots

TRAJECTORY

Milestones of relentless curiosity.

A chronicle of government research, industry collaborations, and foundational leadership.

Jan 2026 – April 2026 Autonomous Systems

MARL Tag: Multi-Agent Pursuit-Evasion Lead

Independent Research Initiative

Architected 3v3 continuous self-play pursuit-evasion framework in PyTorch. Scaled across 8 CUDA subprocess environments achieving 95% victory rate over 30M timesteps.

PPOGAERaycastingPymunk

June 2025 – July 2025 Government Fellowship

AI and Security Research Intern

Ministry of Electronics and Information Technology (MeitY)

Engineered aesthetic error-corrected QR steganography utilizing Stable Diffusion & ControlNet. Pioneered dual-value QR architectures with distance-gated cryptographic payloads for anti-counterfeit access control.

ControlNetSteganographyCryptographic Verification

Jan 2025 – April 2025 Distributed Simulation

Multi-Agent Vehicle Fleet Optimization

Autonomous Dispatch Research

Created an intelligent urban dispatch engine integrating A* pathfinding and MADDPG multi-agent actors, slashing passenger transit delay by 20% across high-density simulated grids.

MADDPGA* SearchFleet Coordination

July 2024 – Aug 2024 Industry Internship

AI & Cloud Intern

Edunet Foundation in collaboration with IBM

Designed an automated data validation and ARIMA predictive forecasting pipeline, deploying model endpoints onto scalable IBM Cloud services.

IBM CloudARIMAPredictive Analytics

Aug 2022 – May 2026 Degree & Leadership

B.Tech in Information Technology

Manipal University Jaipur (MUJ)

Head of Projects & Research at ACM MUJ Student Chapter (recruited and led 15+ student developers). Core research contributor to CyberSpace club. Rigorous coursework in DSA, OS, RDBMS, and Web Tech.

ACM HeadCyberSpaceDSAOS

AG

EXECUTIVE CANDIDATE DOSSIER

AI / Reinforcement Learning & Systems Engineer

Delhi / Jaipur, India · Open to Remote & Global Relocation

30,000,000 PPO Training Timesteps

95.2% MARL Win Rate

MeitY & IBM Research Credentials

May 2026 Graduation Available

aryangupta.2k5@gmail.com +91-9474584883 GitHub LinkedIn

Download Official PDF Resume

Aryan Gupta. Supercharged for Intelligence.