Projects &
Research
17 projects across 7 years — spanning published research, live products, ML systems, and hardware builds.
Showing 17 of 17 projects
Personalized Preparation in Early Autism Intervention
Clinical Trial · Maya Care and Grow · Agartala

Hierarchical Bayesian adaptive N-of-1 randomized crossover trial identifying the optimal pre-session warm-up (Stimulating, Calming, Child-Led, or None) for each child with autism.
Compass — Platform for Intentional Human Connections
Founder & Maintainer

Free, open-source platform to help people form deep connections — platonic, romantic, or collaborative. Keyword search, transparent database, no ads, no hidden algorithms.
Greedy Decentralized Auction-based Task Allocation
Graduate Research · UT Austin

Greedy Coalition Auction Algorithm (GCAA) assigning hundreds of agents to thousands of tasks in real time — applications in aerial firefighting, ride-sharing, and warehouse robotics. One task assigment per auction phase; faster than traditional auction algorithms.
Design of a Smart Sensor for Bird Detection
Master's Thesis · UCLouvain

Ultra-low-power audio IoT sensor harvesting solar energy via supercapacitor. Runs a KNN classifier on-device to discriminate four Belgian bird species at 94% precision. 15+ year lifetime, LoRaWAN communication.
Design of a Wheeled Driving Robot
Project in Mechatronics · UCLouvain · Eurobot

Fully autonomous robot for the Eurobot contest. Implemented path planning (potential fields), Kalman-filtered LIDAR localisation, low-level wheel control, and an Android Wi-Fi remote — running on Raspberry Pi + FPGA.
Large Language Models from Scratch
Personal Project

Full transformer implementation including encoding, embedding, multi-head attention, and MLP layers. Trains an LLM of any size on arbitrary text, and fine-tunes GPT-2 on custom corpora for text generation.
Convolutional Neural Network for Digit Recognition
ML / AI Series #1

Handwritten digit recognition using a CNN trained on the MNIST dataset. The model processes images through feature extraction and a multilayer perceptron to output the most likely digit.
Multi-Agent Motion Planning with Military Maps
Graduate Research · UT Austin · Army Research Lab

Dijkstra-based path planning over military terrain — obstacles, elevation, and valence maps. Runs a fast global route over a downsampled mesh, then refines only the next executed segment at full resolution, cutting compute time significantly without sacrificing local path quality.
Covariance Steering Games with Wasserstein Distance
Project in Multi-Agent Systems · UT Austin

Two algorithms for discrete-time linear covariance steering dynamic games. Iterative best response and LQG reformulation via Riccati equations, evaluated on convergence and solution quality.
Prediction of Air Quality in Beijing
Project in Machine Learning · UCLouvain

PM2.5 concentration prediction using regression models on meteorological data. Covers feature selection (Mutual Information), extraction (PCA), error estimation (Bootstrap 632), and Neural Network / KNN / Lasso / tree models.






