Aarya Gupta

MIDAS: Multimodal Digital Media Analysis Lab IIITD

Computer Vision Intern • Aug, 2024 — Present

Key Responsibilities and Achievements Include:

• Dataset Curation & Preprocessing: Reorganized and standardized a multimodal traffic-accident dataset of 5,691 videos (3,912 train, 1,054 val, 725 test) across CCTV, dashcam, and drone sources—downsampled to 15 fps, resized to 224×224, and structured into overlapping 16-frame clips to enable robust model training and evaluation.
• Dynamic Vehicle Tracking Pipeline: Engineered a hybrid YOLO/DINO + SAM2 segmentation pipeline that dynamically detects and adds newly introduced vehicles into the tracker, propagating masks across subsequent frames via a custom “SAM-2 Alpha” propagation step—validated over thousands of frames for continuous, robust tracking in live traffic streams.
• Real-Time Stream Processing & LLM Verification: Developed a near real-time pipeline processing incoming video in 100-frame (10 s at 10 fps) windows—masking vehicles via CV+SAM2, flagging critical frames by bounding-box overlap, and querying a fine-tuned multimodal LLM—achieving > 90 % video-level accident classification accuracy.
• Advanced Fusion & Temporal Modeling (Evaluation): Integrated contrastive-trained attention fusion across RGB, segmentation, and depth tokens with a Transformer+GRU temporal head, delivering strong frame-level Average Precision and low Mean Time to Anticipation (mTTA) across benchmarks such as DAD, CCD, A3D, and DADA-2000.

NETSEC: Network Security Lab IIITD

Cyber Security Intern • Aug, 2024 — Present

Key Responsibilities and Achievements Include:

• Developed an end-to-end system to auto-evaluate mobile apps’ adherence to declared privacy policies by integrating device-integrity checks, stealthy root bypasses, real-time data capture, and AI-driven policy-vs-behavior analysis.
• Designed and built a dummy Android app leveraging the Google Play Integrity API to verify device tampering status.
• Employed Magisk and KernelSU at the kernel level to stealthily bypass root-detection while keeping the device “pristine.”
• Hooked into running apps with Frida to intercept and log all network traffic (HTTP/S and other protocols) without triggering any alerts.
• Parsed intercepted payloads—device IDs, location data, messages, contacts—and securely logged metadata for forensic analysis.
• Integrated a large language model to compare each app’s declared privacy policy against actual transmission logs, automatically flagging undocumented data leaks.
• Orchestrated a fully automated pipeline for bulk-analysis; from integrity checks and root bypass to dynamic instrumentation and AI-powered compliance reporting.
• Delivered a scalable, repeatable framework that not only bypasses Play Integrity checks undetected but also generates detailed compliance reports—categorizing apps as compliant, partially compliant, or non-compliant—to empower developers, regulators, and privacy advocates.

NAS: Networked Autonomous Systems Lab IIITD

Data Science Intern • May, 2024 — Aug, 2024

Key Responsibilities and Achievements Include:

• Investigated Uncertainty in the autonomous vehicle motion to ensure safety and reliability.
• Developed Probabilistic and Statistical models in Python to forecast vehicle trajectories, explored 3 different approaches (Bayesian Updation, Kernel Density Estimation, Bootstrapping) to optimize predictions by modifying/estimating various degrees of freedom based on the available data.
• Focused on identifying the best confidence interval for the true mean of the underlying distribution using the BCA method after evaluating 5 approaches.

Indraprastha Institute of Information Technology, Delhi

B.Tech Electronics and Communication Engineering • 2022 — present

Salwan Public School Mayur Vihar

CBSE (Class 12th) • 2021 — 2022

96.8%

High-Performance Trade Simulator

Sole Developer • May 2025 — Jun 2025

• Built a Python L2 order-book simulator for OKX, ingesting live WebSocket ticks (~10 Hz) and driving a multithreaded Tkinter UI with sub-10 ms per-tick latency.
• Modeled transaction costs via slippage regression on 100k+ synthetic samples (Test R² > 0.6), rule-based fees, and a simplified Almgren–Chriss impact model.
• Engineered a multithreaded architecture, logging and visualizing model performance metrics (MSE, R²) to guide iterative refinements.

PerSpectraMed – Perspective-Aware Summarization for Healthcare QA

Research Engineer • Jan 2025 — May 2025

• Implemented a multi-perspective QA summarization pipeline by fine-tuning Gemma2 via LoRA with a custom span-prediction head on the PUMA CQA dataset.
• Benchmarked against Gemini 2.0 Flash, Gemini 3 27B, PLASMA, and other baselines, achieving superior ROUGE-L and BERTScore.
• Built a reproducible evaluation suite computing ROUGE, BLEU, METEOR, and BERTScore to quantify factuality and compression gains.

IntelliDoc – AI-Powered Health Report Interpreter

Full-Stack Developer • Jan 2025 — May 2025

• Built a full-stack AI web platform for uploading and interpreting medical reports and images (X-rays, MRIs), delivering real-time summaries, condition predictions, and specialist/medication recommendations.
• Included a layman-friendly medical chatbot backed by Node.js, MongoDB, Flask, and TensorFlow.
• Architected a freemium B2C SaaS backend and ReactJS/Tailwind CSS frontend with JWT/OAuth2 security and Cloudinary/Multer/OpenCV media handling.

SignTrack – Real-Time Traffic Sign Detection using YOLOv8

Machine Learning Engineer • Jan 2025 — May 2025

• Fine-tuned YOLOv8 on 4.9k+ traffic sign images (15 classes), achieving 95.7% mAP@0.5, 94.2% precision, and 2.4 ms inference latency per image.
• Developed a Flask backend and React frontend with drag-and-drop uploads for real-time detection; containerized with Docker and automated CI/CD via GitHub Actions.

Stick Hero Game

Lead Developer • Aug 2023 — Jul 2024

• Developed a Java-based action platformer with five screens, featuring rewards collection, revive/pause, and save/load via serialization.
• Designed UML and class diagrams; applied design patterns, OOP principles, and modular coding for extensibility.
• Built the UI in JavaFX/SceneBuilder and validated functionality with comprehensive JUnit testing.

Baazaar – E-Commerce Website

Solo Developer • Aug 2024 — Dec 2024

• Crafted a responsive e-commerce platform with HTML, CSS, Node.js, and MySQL supporting 10 concurrent users in stress tests.
• Implemented user registration, login (5 req/sec), shopping cart, and order history modules.
• Conducted functional and performance testing, resolving 10+ issues and boosting reliability by 50%.

Euphoria – A Well Being App

Lead Developer • Nov 2022 — Feb 2023

• Conceived and prototyped an app to connect individuals facing mental health challenges with resources, communities, and tools.
• Led UX research: crafted personas, empathy maps, and developed lo-fi and hi-fi prototypes using Figma and Miro.

Expertise Area

Data Structures and Algorithms, Object Oriented Programming, Competitive Programming, Deep Learning (NLP, CV)

TechStack

Python, Java, C/C++, MySQL, Javascript, MATLAB, Git/GitHub, Linux, VS Code, Cursor, IntelliJ IDEA, Android Studio, Google Colab, Kaggle, HTML/CSS, MySQL, Python Libraries (NumPy, Pandas, Matplotlib, Scikit learn, PyTorch, TensorFlow, OpenCV), Hugging Face, AI Agents, Retrieval Augmented Generation (RAG), CARLA Simulator, Figma, LaTeX, Wireshark

Soft Skills

Effective Communication, Adaptability and Teamwork

Scholarship Holder (Merit Based)

Reliance Foundation • 2023

Awarded the prestigious Reliance Foundation Scholarship post qualifying through a rigourous entrance process.

Electrivire Winner

ECELabs IIITD • 2024

1st position out of 44 competitive teams.

School Topper

Salwan Public School Mayur Vihar Delhi • 2022

Secured 1st position in the Class XII CBSE Board Exams (AISSCE) among a batch of over 250 students.

Undergraduate Research Club IIITD

Researcher • 2024 — present

Reliance Foundation

Scholar Mentor @ Scholar Community • 2024 — present

Placement Cell IIITD

Senior Member • 2024 — present

Reliance Foundation

Volunteer in Workshops • 2023 — present

• Github
• LinkedIn