Frequently Asked Questions

LARUN (Light Analysis Research Utility Network) is an AI-powered platform for discovering exoplanets in NASA telescope data. It uses a federation of TinyML models to detect transit signals - the tiny dips in starlight caused by planets passing in front of their host stars.

Unlike traditional astronomy software, LARUN uses a conversational interface. You can ask questions in plain English, and LARUN handles the complex data processing automatically.

No! LARUN is designed to be accessible to everyone. Our motto is "No PhD required." The chat interface uses natural language, so you can simply ask things like "Search for transits in TIC 307210830" without knowing any programming.

That said, understanding basic concepts like what a transit is will help you interpret results. Check out our User Guide and the ASTRA educational module for learning resources.

Yes! Citizen scientists have contributed to numerous exoplanet discoveries. Notable examples include:

• K2-138 system - discovered through the Exoplanet Explorers project
• TOI-700 d - one of the first Earth-sized habitable zone planets from TESS
• Kepler-90i - found using machine learning by citizen scientists

If LARUN identifies a promising candidate, you can submit it to the TESS Follow-up Observing Program (TFOP) for professional validation.

LARUN offers a free tier that allows you to analyze up to 10 targets per month. This is enough for casual exploration and learning.

For more intensive use, we offer paid plans:

• Explorer ($9/mo): 100 targets, basic features
• Pro ($29/mo): 500 targets, API access, priority support
• Scientist ($99/mo): Unlimited, batch processing, Kepler + JWST data

See our Pricing page for details.

A transit occurs when a planet passes between its host star and Earth, causing a small, temporary decrease in the star's brightness. This dip is typically 0.01% to 2% of the star's light, depending on the planet's size.

By measuring these periodic dips, we can determine:

• Orbital period: How long the planet takes to orbit
• Planet size: Larger planets block more light
• Orbital distance: Combined with stellar properties

LARUN's TinyML transit detection model achieves 81.8% accuracy on our test set. This means:

• ~82% of real transits are correctly identified (true positives)
• ~18% of real transits may be missed (false negatives)
• Some non-transit signals may be flagged (false positives)

To improve reliability, LARUN uses multiple vetting steps including BLS periodogram, model ensemble voting, and false positive probability calculations.

FPP is the probability that a transit-like signal is caused by something other than a planet, such as an eclipsing binary star or instrumental artifact.

• FPP < 1%: VALIDATED - Very likely a real planet
• FPP 1-50%: CANDIDATE - Promising but needs more analysis
• FPP > 50%: LIKELY_FP - Probably a false positive

LARUN calculates FPP by comparing the likelihood of different scenarios (planet, eclipsing binary, background binary, etc.) based on the signal properties.

By default, LARUN searches for periods between 0.5 and 15 days. This range catches most transiting planets observable by TESS.

For longer periods, you can adjust the search range:

"Search for long-period transits up to 30 days"

Note: Detecting long-period planets requires more observation time - at least 3 complete transits for confident detection.

The smallest detectable planet depends on:

• Star size: Smaller stars = deeper transits for same planet
• Data quality: Less noise = smaller signals detectable
• Number of transits: More transits = better signal averaging

For bright, quiet M-dwarf stars with TESS 2-minute cadence data, LARUN can potentially detect planets as small as 1.5 Earth radii. For G-type stars like the Sun, the limit is around 2-3 Earth radii.

TinyML is machine learning optimized for "edge" devices - small, low-power systems like microcontrollers, smartphones, or browsers. LARUN's models are:

• Small: Under 100KB each (fits in browser cache)
• Fast: Inference in under 50 milliseconds
• Efficient: Runs without GPU or cloud servers

This allows LARUN to run entirely in your browser, protecting your privacy and enabling offline analysis.

LARUN uses a federation of 9 specialized models:

• EXOPLANET-001: Transit detection (core)
• VSTAR-001: Variable star classification
• FLARE-001: Stellar flare detection
• ASTERO-001: Asteroseismology analysis
• SPECTYPE-001: Spectral type classification
• GALAXY-001: Galaxy morphology
• SUPERNOVA-001: Supernova classification
• MICROLENS-001: Microlensing detection
• TRAINER-001 (ASTRA): Educational AI tutor

See our Models page for detailed documentation.

LARUN models are trained on real astronomical data:

• Training data: NASA TESS and Kepler light curves
• Labels: Confirmed planets, known eclipsing binaries, etc.
• Augmentation: Synthetic transits injected into real data
• Validation: Tested against held-out confirmed planets

Models are retrained weekly with new data and undergo quality gates before deployment (accuracy ≥80%, size ≤100KB, inference ≤50ms).

ASTRA (Astronomical Science Tutor and Research Assistant) is LARUN's AI-powered educational companion. Unlike the detection models, ASTRA uses Retrieval-Augmented Generation (RAG) to answer questions about space science.

ASTRA can:

• Explain astronomical concepts in plain language
• Provide interactive lessons on exoplanets, stars, and missions
• Give quizzes to test your knowledge
• Explain why models made specific predictions

Try it by typing /learn in the chat!

LARUN fetches data from NASA's Mikulski Archive for Space Telescopes (MAST):

• TESS: Full-frame images, 2-minute cadence, 20-second cadence
• Kepler: Long and short cadence (Scientist tier)
• K2: Extended Kepler mission data
• Gaia DR3: Stellar parameters

All data is publicly available through NASA's open access policy.

Yes! LARUN accepts custom light curves in several formats:

• FITS: Standard astronomical format
• CSV: Comma-separated values
• TXT: Space-separated text

Your data should include time (BJD or relative), flux, and optionally flux errors. See the User Guide for format details.

TIC stands for TESS Input Catalog. Every star observed by TESS has a unique TIC ID number (e.g., TIC 307210830). You can find TIC IDs through:

• ExoFOP-TESS
• MAST Portal
• Searching by coordinates or common name in LARUN

API access is available on Pro and Scientist tiers. To get your key:

1. Sign in to your account
2. Go to Settings → API Keys
3. Click "Generate New Key"
4. Copy and store securely (it's only shown once)

Yes! We offer educational discounts for students and researchers. Contact us at academic@larun.space with your institutional email for special pricing.

If you publish research using LARUN, please cite us:

LARUN.SPACE, 2026, "TinyML Federation for Exoplanet Detection"
https://larun.space

When you reach your monthly target limit, you'll receive a notification. You can:

• Wait until next month (limits reset on billing date)
• Upgrade to a higher tier
• Purchase additional targets as a one-time add-on

We never charge overage fees without your explicit consent.

Partially. The TinyML models run entirely in your browser and can work offline once loaded. However, you need internet access to:

• Fetch light curves from NASA MAST
• Use ASTRA (requires LLM API)
• Save analyses to cloud storage

For fully offline work, download your target data in advance.

LARUN works on modern browsers that support WebGL and WebAssembly:

• Chrome 90+ (recommended)
• Firefox 88+
• Safari 14+
• Edge 90+

For best performance, use Chrome with hardware acceleration enabled.

Yes. LARUN processes your data locally in your browser. We don't upload your light curves or analysis results to our servers unless you explicitly choose to save them.

If you use ASTRA, your questions are sent to the LLM provider (Anthropic, OpenAI, or Ollama depending on configuration), subject to their privacy policies.

Please report bugs through:

• GitHub Issues (preferred - helps other users too)
• Email: support@larun.space

When reporting, please include:

• What you were trying to do
• What happened instead
• Browser and OS version
• Target ID if applicable

Yes! LARUN is open source. You can deploy it yourself:

git clone https://github.com/Paddy1981/larun.git
cd larun
pip install -r requirements.txt
python larun.py

See the GitHub repository for full setup instructions.