Research & Projects

1st Contact AI is an AI-powered classification tool that automates the vetting of potential exoplanet transit signals from NASA’s Kepler, K2, and TESS missions. By distinguishing true planetary candidates from common false positives (such as eclipsing binaries, variable stars, or instrumental noise), the tool significantly speeds up and improves the reliability of exoplanet discovery.

I chose to pursue this project during the 2025 NASA International Space Apps Challenge because it perfectly combined my growing interest in Generative AI with real scientific impact. As someone exploring AI applications beyond financial services, I wanted to apply modern machine learning techniques to open NASA datasets and contribute to one of humanity’s most inspiring quests: finding worlds beyond our solar system. The project was inspired by Carl Sagan’s Contact — reimagining “first contact” as AI helping humanity detect and prioritize new exoplanets.

I led a talented global team of developers, data scientists, and space enthusiasts. The project was completed over the intense 48-hour hackathon weekend (October 4–5, 2025) and earned Global Nominee status — one of the highest recognitions from thousands of submissions worldwide.

Generative AI tools (ChatGPT, Gemini, Grok, Ollama, Midjourney) were used ethically for ideation, coding assistance, debugging, and visuals.

Objectives

The initial situation was clear: NASA’s exoplanet missions generate enormous volumes of light-curve data, but identifying genuine planets among the noise requires extensive manual review by astronomers. This bottleneck slows discovery, introduces human error, and limits access for students and citizen scientists.

Our primary objectives were:

Build a fast, accurate, and accessible AI model to classify transit signals as Planet, Eclipsing Binary, Variable Star, or Instrumental Noise.

Reduce manual vetting time while improving consistency and minimizing false positives/negatives.

Make NASA’s open data more usable by creating an intuitive interface with visualizations, confidence scores, and downloadable results.

Handle real-world challenges such as severe class imbalance in the dataset and ensure reproducibility.

Explore optional LLM-based explainability to make model decisions more transparent and interpretable.

Key considerations throughout included using only publicly available NASA data, keeping the tool lightweight and deployable (via Streamlit on Hugging Face), supporting multiple ML backends for flexibility (LightGBM, XGBoost, CatBoost, Random Forest), and maintaining a strong focus on open science and accessibility.

Results

The project successfully delivered a working, deployable tool that automates exoplanet transit classification with strong performance metrics. As team lead, I built the initial prototype (feature engineering, model training, and pipeline) which the global team then refined into a polished solution.

Key outcomes:

Achieved Global Nominee recognition in the “A World Away: Hunting for Exoplanets with AI” challenge.

Users can upload light-curve feature CSVs and receive instant predictions with confidence scores, confusion matrices, and visualizations.

The switchable ML backends allow performance comparison, while an optional LLM layer adds explainability.

The tool helps accelerate exoplanet cataloging and supports downstream efforts like prioritizing habitable-zone candidates for SETI and atmospheric studies (e.g., with JWST).

Unexpected highlights included the rapid global collaboration and the project’s timely cultural resonance — aligning with growing public and policy interest in the search for extraterrestrial life in early 2026.

What I learned: Leading under tight hackathon constraints reinforced the power of “vibe coding” to create momentum quickly, the importance of handling class imbalance in scientific datasets, and how Generative AI can dramatically speed up prototyping without replacing core engineering judgment.

The results have clear future impact: the open-source components and Hugging Face deployment make the tool extensible for researchers. It lays groundwork for integrating with broader astrophysics and SETI workflows, helping prioritize promising exoplanets for deeper observation.

This project strengthened my confidence in applying AI to complex, real-world scientific problems and demonstrated how skills from fintech tech can translate into meaningful contributions in other domains.