What I Built

Most modern astronomy and weather applications do the opposite of "touching grass": they flood users with infinite radar loops, predictive maps, and engagement notifications—keeping eyes glued to bright blue-light screens right before bedtime.

AstroCast was built around a singular design rule: make the screen the shortest part of the experience (under 30 seconds).

AstroCast is an offline-first atmospheric seeing predictor and celestial observation scout for amateur astronomers, campers, and anyone wanting to step outside into the night air:

One Tabular Quality Score (0–100%): Rather than asking users to interpret complex raw meteorological data, it evaluates atmospheric turbulence (Seeing) and photometric light transmission (Transparency) into a single Stargazing Quality Index.

Prime Celestial Window: Calculates the exact time envelope tonight when the sky will be darkest and most stable before fog, humidity rise, or moonrise.

Naked-Eye Targets: Recommends 3–4 visible wonders (e.g., Jupiter rising in Taurus, Saturn in Aquarius, the Pleiades cluster, and the Andromeda Galaxy) that require zero telescopes.

The Screen Cutoff Enforcer: Features an integrated 30-second dark-adaptation countdown. Human rod cells require ~20 minutes in real darkness to synthesize rhodopsin; AstroCast gives you your target, then actively urges you to shut off your device and step outside.

Field Red-Light Mode: One-click toggle transforms the entire interface into 650nm monochrome dark-crimson on obsidian black—the standard light spectrum used by astronomers in field tents to preserve night adaptation.

Code

titas-mahato / astrocast

ASTROCAST ✦

Gaze into the cosmos. Built on open-weight AI (TabPFN architecture) for night sky forecasts.

Overview

Most modern astronomy and weather applications flood users with infinite satellite maps, ads, and engagement notifications—keeping eyes glued to blue-light screens right before bedtime.

AstroCast is built with a singular design principle: minimize screen time to under 30 seconds.

Select your location, and AstroCast's open-source tabular AI model evaluates atmospheric turbulence (seeing) and clarity (transparency) to output a single Stargazing Quality Index (0–100%), an optimal observing window, and 3 naked-eye celestial wonders. It then actively prompts you to shut off your screen so your eyes can adapt to real-world darkness.

Key Features

Open-Weight Tabular AI Core: Uses a physics-grounded tabular gradient boosting pipeline (TabPFN architecture) evaluating microclimate indicators (dew point depression, boundary layer wind shear, barometric variation, cloud stratification, and lunar phase).
Active Screen Cutoff ("Touch Grass" Mode): Features…

View on GitHub

How I Built It

Open-Source Tabular AI Core: Rather than using bloated closed LLMs that hallucinate atmospheric physics, AstroCast uses an open-weight tabular machine learning pipeline (TabPFN architecture + Gradient Boosting) trained on microclimate atmospheric vectors.

Atmospheric Seeing Physics:

Dew Point Depression (Temp - Dew Point): Measures proximity to saturation; a gap under 2°C indicates a high risk of ground fog and optical mist.
Boundary-Layer Wind Shear: Analyzes 10m surface winds to quantify mechanical air turbulence that degrades seeing resolution (FWHM).
Stratified Cloud Column: Weighs low-altitude versus high-altitude cirrus clouds against atmospheric transmission.

Conway Lunar Algorithm: Computes real-time synodic lunar phases and illumination percentages to calculate skyglow interference.

Keyless Meteorological Pipeline: Built with Python FastAPI, pulling live numerical weather model data from Open-Meteo (open-source, zero API keys required).

Autonomous Offline Fallback: If taken deep into remote mountains with zero cell coverage, AstroCast automatically falls back to an internal microclimate physics simulator so the app never fails on the trail.

Why Does Open Innovation Matter?

Trail Independence (100% Offline): The best stargazing happens miles away from cellular networks in high-altitude sanctuaries like Hanle (Ladakh) or Spiti Valley. Because the AI model and astronomical algorithms run on local open inference, AstroCast runs autonomously on a laptop with zero internet.

Tabular AI vs. Generative Chatbots: Tabular foundation models like TabPFN prove that open AI can solve structured physical and environmental prediction tasks reliably without burning kilowatts on multi-billion parameter LLMs.

Intentional Screen Minimization: Commercial ad-driven apps design for endless engagement. Open-source innovation allowed us to build an app whose primary goal is telling you to put your phone in your pocket and look up at the stars.

DEV Community

AstroCast: The Celestial Window