Using AI and technology to create real-world solutions for social good

All projects are built, deployed, or published by Ansh Gupta, Founder - a high school researcher working at the intersection of human-centered AI and social impact.


Research & Impact Projects

Project 1. Project Refluent

Refluent is the first personalized AI communication tool designed specifically for stroke survivors with Non-Fluent Aphasia. Built using a fine-tuned transformer-based NLP model, it reconstructs fragmented speech into coherent text or audio in real time – preserving the user's intent. Developed in collaboration with speech pathologists at SPARC Health.

Status: Currently piloting with 100 patients across partner hospitals in New Delhi to learn from user feedback.
Model Scores: BLEU 41.24 | ROUGE1 75.57 | 235+ HuggingFace downloads

Project 2. ML Framework for Drug Overdose Prevention

A scalable machine learning framework that predicts drug overdose hotspots and guides public health resource allocation. Built using XGBoost and Tweedie GLM on CDC and US Census data for King County, WA – cross-validated on Los Angeles data to prove national replicability. The framework introduces a Budget Prioritization Rate (BPR) to translate predictive outputs directly into policy decisions, achieving 97.6% allocation efficiency.

Recognized with 2nd Place at both the Central Sound and WA State Regional Science & Engineering Fairs (2026).

Status: First-author manuscript in progress. Conversations with WA state senate in progress.
Model: Open source on HuggingFace

Project 3. ASL Avatar

An LSTM-based model that converts spoken audio into real-time digital representations of American Sign Language. This project led to a 10-page technical proposal submitted to the MIT THINK program, outlining an enhanced FSQ-based representational learning approach to capture nuances in hand shape and facial expression.

Status: This project is currently on pause while we seek domain expertise and mentorship in representational learning. A 10-page technical proposal outlining the next phase was submitted to the MIT THINK program. We are actively looking for research collaborators to take this to the next level.

Project 4. Safe Circle: Anonymous Support for Students

Safe Circle is a web-based platform that gives students a safe, anonymous space to share academic or personal challenges, including concerns about bullying. AI-powered analysis helps categorize reports, detect patterns, and provide schools with meaningful insights to strengthen their support systems.

Goal: Promote student safety, mental well-being, and early intervention.