Aashna's Portfolio
AI-mmunity
AI-mmunity is an AI-powered public health platform that helps both communities and health officials detect, report, and respond to real-time risks like pollution, illness trends, or safety incidents. Built with multi-agent workflows, live datasets, and crowdsourced reporting.
Winner of Google’s Agentic AI Arena Hackathon 2025.
Parallell Local Neighbourhood Difference Pattern Feature Extractor (CUDA based)
Co-authored a paper on ⍬(1) time complexity algorithm for Local Neighbourhood Difference Pattern (LNDP) feature extraction, accelerating medical image analysis using GPU-based CUDA implementation. Tested on the LISS medical image dataset, achieving massive speedups for feature computation in diagnostic applications like epilepsy detection.
Spark-Based NLP Pipeline for Safe LLM Outputs
Designed a distributed text analysis pipeline using Apache SparkML to detect hate speech and analyze sentiment in large-scale GPT outputs. Built an end-to-end workflow for data generation, preprocessing, and model training to improve the safety of AI-generated content.
Real-time Parkinson's Tremor detection system
Built a wearable device with STM32 and gyroscopic sensors to detect and classify Parkinsonian tremors in real time, providing instant feedback and continuous motion data tracking.
RISC-V 5-Stage Pipeline Simulator
Register-based simulation of a 5-stage pipelined RISC-V processor, focusing on instruction throughput optimization and hazard mitigation techniques. This project serves as an educational tool for understanding CPU architecture and pipeline processing.
Using GANs to generate CBOE volatility index values
Employed Generative Adversarial Networks (GANs) to model and generate synthetic data replicating the Chicago Board Options Exchange (CBOE) Volatility Index (VIX). This project provides insights into financial market simulations and risk assessment!
Résumé.sh
Built a fun, interactive, command-line style résumé. The terminal UI supports real commands (e.g., help, resume, projects), allowing recruiters to explore one's background as if they're inside a live shell. Supports command history, shell-style prompts, and prompt-driven navigation.
Try it! :)
Predictive QoS for V2X Networks with Machine Learning
Developed an ML-based system that predicts mobile network performance for moving vehicles. Using real-world V2X data (signal metrics, GPS, speed, and environment), the model learns patterns in connectivity and forecasts future network quality. This helps fleets and smart vehicles anticipate low-signal zones, optimize routing, and maintain stable data flow during travel.
Optimized ResNet Architecture for Image Classification with Squeeze-and-Excitation Blocks
Built and trained a custom ResNet with Squeeze-and-Excitation blocks for image classification on the CIFAR-10 dataset, using advanced data augmentation and gradient optimization techniques to improve generalization and model performance.
"Bachman"
Blended music theory and retro gaming! Bachman is a playable Pac-Man-style game designed to teach and reinforce music note recognition. Built in Lua using LÖVE2D on macOS and hosted on Firebase.
AI-Driven PCAP Flow Analyzer for Network Vulnerability Detection
A web-based AI tool that inspects .pcap files and classifies network flows as normal or suspicious using LLMs. Combines packet-level parsing with GPT-based reasoning to surface potential exfiltration or anomaly patterns in real time.
File Read Performance Benchmarking using C and Google Benchmark
Designed and implemented a performance benchmarking framework in C/C++ to evaluate cached vs. non-cached file reads across varying block sizes using both custom timing functions and the Google Benchmark library.
Neural Network from Scratch (Backpropagation in PyTorch & NumPy)
Implemented a simple feedforward neural network entirely from first principles by manually coding the forward and backward propagation steps using NumPy before verifying with PyTorch.
Helps visualize how gradient descent and weight updates work under the hood.