🌌 What I Built

1. The Biological Paradox of Modern Stargazing Every popular stargazing application on the market makes the exact same catastrophic design assumption: it asks you to hold a radiant 6.1-inch OLED screen 30 centimeters from your eyes while standing outside in the dark. Here is the neurological and biochemical reality: The human retina relies on rhodopsin (visual purple), a sensitive photopigment concentrated within 120 million rod cells, to perceive low-luminance deep-sky structures—the gossamer spiral arms of the Andromeda Galaxy, the ghostly veil of the Milky Way rift, or subtle emission nebulae. When your eyes are exposed to the glare of a smartphone display—even on the lowest brightness setting or filtered through artificial night modes—the photons instantly bleach rhodopsin back into retinal and opsin. It takes between 20 and 30 continuous minutes of absolute darkness for rod cells to regenerate this pigment and restore full scotopic (dark-adapted) vision. Every single time a user checks a celestial map, tweaks a slider, or reads an on-screen notification, their dark adaptation is destroyed. Traditional astronomy software literally blinds you to the very stars you stepped outside to observe. --- ### 2. Enter Whispers of the Sky (WOTS) Whispers of the Sky (WOTS) answers the eternal human inquiry: "WOT's in the sky tonight?" WOTS is a screenless, audio-first celestial companion engineered with open-source AI at its core. It operates under a radical human-computer interaction thesis: the screen must be the shortest interaction in the experience—under 30 seconds total. Instead of gazing down at glass pixels, the observer lies horizontally on grass or a wool blanket with their phone resting face-down upon their chest. At that moment, the device completely shuts off its display to true 0-nit OLED blackness, shifts interaction to subtle haptics and whispers, and conducts a 100% real-time, body-aligned audio tour of the living universe.

3. Core Architectural Capabilities

A. 🖐️ Blind Physical Gestures (No Looking at Screen)

When lying in the dark outdoors, you should never have to illuminate your screen to control playback. WOTS implements tactile, sensor-driven blind gestures:

• Single Tap on Phone Back: Toggles Pause / Resume. The phone emits a subtle 50ms haptic pulse and whispers confirmation: "Paused." or "Resumed."
• Double Tap on Phone Back: Skips immediately to the next live celestial target (e.g., advancing from Polaris to Saturn) accompanied by a distinct double-pulse haptic vibration.
• Lifting the Phone (Face-Up): Dual accelerometer and gyroscopic Euler tracking detect when the phone is raised or tilted up. Narration gently pauses with an audio bookmark: "Holding your place in the sky."
• Resting Phone Face-Down on Chest: The screen instantaneously blacks out to 0 nits, and the celestial tour smoothly resumes: "Resuming under the stars."

B. 🌌 100% Live Anatomical Sky Tour (Zero Canned Chapters)

WOTS rejects pre-recorded, generic MP3 files. Every single tour is computed dynamically in real-time based on your exact GPS coordinates and device compass heading:

1. Body Alignment & Grounding: Locks your spine’s cardinal heading (Head pointing North, Toes pointing South) with somatic breathwork.
2. Overhead Zenith Target: Identifies what is currently resting directly above your chest and eyes (e.g., The Summer Triangle or Cassiopeia).
3. Crown of Head Horizon: Points backward past your crown toward your northern horizon anchor (Polaris).
4. Feet & Toes Horizon: Sweeps your gaze downward across your toes toward bright planetary beacons (Saturn / Jupiter).
5. Lateral Shoulders: Traces the vast expanse of the galactic rift between your left and right horizons (The Great Square of Pegasus and Andromeda).
6. Deep Cosmic Stillness: Concludes with atmospheric seeing transparency contemplation and peaceful silence.

C. 🎙️ Hands-Free Speech-to-Speech Q&A (Voice In, Voice Out)

While resting in the dark, the observer can speak aloud at any moment:

"Where is Saturn right now?"

"What's resting directly above my head?"

"Tell me the mythology of Cassiopeia."

WOTS detects speech hands-free, pauses narration, chimes a gentle 528 Hz Solfeggio acoustic bell ping, queries the local celestial intelligence backend, and whispers the answer directly into your earbuds via George or Sarah, before resuming the live tour seamlessly.

D. 🎙️ Classy Dual Narrators: George & Sarah

Observers can switch between two world-class narrative personalities:

• George: Warm, deep, contemplative British planetarium guide.
• Sarah: Soft, gentle, whispered nocturnal companion.

E. ☀️ 24-Hour Adaptive Situational Intelligence

Stargazers don't only exist at midnight. WOTS adapts to every phase of an observer's day:

• ☀️ Daytime Mission Planning Mode (Sun Alt > 0°): Computes real-time solar elevation, previews tonight's celestial targets crossing the meridian under sunlight, and highlights the evening's golden seeing window.
• 🌅 Golden Hour & Twilight Transition (-6° ≤ Sun Alt ≤ 0°): Detects civil sunset, gradually primes retinas, and guides eyes to the first beacons piercing dusk (Venus, Jupiter, Vega).
• 🌲 Wilderness Stargaze (Sun Alt < -18°): The core 0-nit deep-sky tour for dark-sky lawns and parks.
• 🌆 City Sky Mode (Urban Light Pollution Filter): Automatically filters out faint nebulae and focuses on high-magnitude urban punchers (Planets, Moon craters, Sirius, Vega) that pierce city streetlights.
• 🌧️ Indoor Cosmic Sanctuary: When rain or heavy overcast clouds roll in, WOTS switches to an indoor sleep sanctuary—synthesizing a soothing 432 Hz rain drone and deep-space bedtime astrophysical lore.
• ⏳ Smart Sleep Timer: Observers drifting off to sleep can engage a 15m, 30m, or 45m timer that gently fades volume to zero and halts playback to conserve phone battery.

🎥 Demo

• Live Global Deployment: https://whispers-of-the-sky.onrender.com/
• GitHub Repository: https://github.com/FrozenLionMax/whispers-of-the-sky

Interactive UI Highlights

• VisionOS Astrolabe Celestial Orb: One-tap master experience launcher with continuous audio-visual state syncing.
• Real-Time Anatomical Telemetry HUD: Live Euler pitch, roll, and gravity-Z accelerometer sensors detecting chest placement.
• 650nm Monochromatic Astronomical Red Mode: Hardware-level blue-light elimination for setup before screenless mode.
• Hemispheric Sky Radar Dome: Interactive 2D projection of visible targets mapped to altitude rings and azimuth compass bearings.

💻 Code Architecture

The complete open-source codebase is hosted on GitHub:

System Architecture Pipeline

flowchart TD subgraph Physical World & Observer User["Observer Lying on Grass (Headphones On)"] ChestSensor["Chest Orientation Sensors (Gravity Z & Euler Beta/Gamma)"] NightSky["Living Celestial Heavens (Andromeda, Saturn, Vega, Polaris)"] end subgraph WOTS Multi-Model AI Engine Stack TabPFN["Prior Labs TabPFN\n(Micro-Meteorological Tabular Seeing Forecaster)"] Astropy["Astropy Celestial Mechanics\n(Real-Time Alt/Az & Solar Elevation Engine)"] Gemma["Google Gemma 2B via Local Ollama\n(Edge Celestial Reasoning & Voice Q&A)"] ElevenLabs["ElevenLabs Neural Voice AI\n(Low-Timbre Whispered Synthesis & Cache)"] WebAudio["Procedural Web Audio Engine\n(432Hz Cosmic Drone & 528Hz Solfeggio Chimes)"] end ChestSensor -->|Face-Down 0 Nits / Taps / Tilt| Astropy TabPFN -->|Atmospheric Transparency & FWHM Arcsec| Gemma Astropy -->|Anatomical Alt/Az Targets & Solar Phase| Gemma Gemma -->|Spoken Spatial Narration & Answers| ElevenLabs ElevenLabs -->|Audio Tour Stream & Offline Cache| User WebAudio -->|Harmonic Soundscape| User User -.->|Bare Eyes (Full Scotopic Vision)| NightSky

🔬 How I Built It: Deep Scientific & Technical Theory

Whispers of the Sky weaves together astrophysics, atmospheric fluid dynamics, neural machine learning, and auditory psychoacoustics. Here is the deep theoretical foundation powering each component:

1. Atmospheric Turbulence & Seeing Prediction with Prior Labs' TabPFN

The Physics of Astronomical Seeing

Starlight travels unhindered across billions of light-years of near-perfect vacuum, only to be violently distorted in the final 100 microseconds as it traverses Earth's turbulent atmosphere.

According to Kolmogorov turbulence theory, thermal differentials between atmospheric layers produce refractive index fluctuations ($\Delta n$). These turbulent eddies break coherent optical wavefronts, causing point sources of light (stars) to blur and oscillate. The quality of this optical transmission is quantified as Astronomical Seeing, measured by the Full-Width at Half-Maximum (FWHM) of the stellar point-spread function in arcseconds.

Standard astronomy applications rely on static weather forecasts or simplistic linear heuristics. In reality, seeing is governed by non-linear interactions across micro-meteorological variables:

• Boundary layer temperature differentials ($T - T_{\text{dew}}$)
• Barometric pressure gradients ($P_{\text{hPa}}$)
• Relative humidity ($\text{RH}\%$) and condensation nucleation
• Wind shear velocity ($v_{\text{wind}}$)
• Aerosol optical depth ($\tau_{\text{aerosol}}$)

Why TabPFN is the Optimal Foundation Model

Instead of training brittle classical regressors, WOTS leverages TabPFN (Prior Labs)—a breakthrough tabular foundation model pretrained on synthetic datasets generated from millions of structural causal models and Bayesian neural network priors.

# From core/tabpfn_predictor.py from tabpfn import TabPFNClassifier, TabPFNRegressor class TabPFNSeeingPredictor: def __init__(self): # Ingests micro-meteorological variables to forecast FWHM and transparency  self.classifier = TabPFNClassifier(device='cpu') self.regressor = TabPFNRegressor(device='cpu') def predict_seeing(self, weather_features: Dict[str, float]) -> Dict[str, Any]: # Evaluates atmospheric transparency index (%) and FWHM arcseconds  # Generates the 8-hour Golden Window observation timeline  ... 

Theoretical Advantages of TabPFN in Outdoor Astronomy:

1. Zero Hyperparameter Tuning: TabPFN performs in-context learning in a single forward pass, predicting exact non-linear seeing thresholds without gradient descent or overfitting.
2. Robust Multi-Task Inference: WOTS simultaneously deploys TabPFNRegressor to forecast the continuous Atmospheric Transparency Percentage and TabPFNClassifier to categorize astronomical seeing tiers (Pristine, Average, Poor).
3. The Golden Hour Timeline: TabPFN models an hourly timeline from 8 PM to 4 AM, identifying the exact moment of minimum atmospheric scintillation so observers know precisely when to look up.

2. Edge Celestial Intelligence with Google Gemma 2B

The Constraint of Wilderness Computing

True dark-sky sites—such as International Dark-Sky Association (IDA) reserves, alpine trails, and coastal cliffs—share a universal feature: total absence of cellular connectivity.

Proprietary cloud LLMs (OpenAI, Anthropic, Gemini API) are fundamentally incapable of operating in the wild. When a network socket fails, cloud-tethered applications display infinite loading spinners and crash.

Open-Weight Inference via Local Gemma

WOTS deploys Google's open-weight Gemma 2B running locally via Ollama (gemma2:2b). Gemma serves as an on-device naturalist, converting raw Astropy horizontal coordinates into evocative, pedagogical guidance:

# From core/gemma_astronomer.py class GemmaAstronomer: def answer_astronomy_question( self, question: str, visible_targets: List[Dict[str, Any]], seeing_data: Dict[str, Any] ) -> str: # Prompting Gemma with current visible Alt/Az context and TabPFN seeing quality  # Fast 6-second timeout with semantic entity fallback ensures zero offline latency  ... 

Cognitive Capabilities of the Gemma Engine:

• Spatial Coordinate Translation: Transforms dry numerical coordinates (e.g., Azimuth 42°, Altitude 58°) into somatic physical instructions ("Tilt your chin back sixty degrees toward the northeast").
• Multi-Cultural Celestial Lore: Synthesizes cross-cultural sky heritage, bridging 964 AD Persian astronomy from Abd al-Rahman al-Sufi's Book of Fixed Stars, Babylonian star-paths, ancient Greek myths, and modern stellar nucleosynthesis.
• Instant Voice Q&A: Employs rapid semantic object matching combined with Gemma inference, answering spoken queries in under two seconds.

3. Auditory Psychoacoustics & Neural Voice with ElevenLabs

The Acoustics of the Nocturnal Outdoors

The ambient sound pressure level of an outdoor wilderness meadow at night drops below 25 dBA. Standard synthetic text-to-speech voices sound jarring, sterile, and metallic, triggering startle reflexes and alerting nocturnal fauna.

WOTS integrates ElevenLabs Neural Voice AI calibrated for midnight listening:

• George (JBFqnCBsd6RMkjVDRZzb): High stability (0.75), low pitch, and deep meditative cadence.
• Sarah (EXAVITQu4vr4xnSDxMaL): Soft, whispered, intimate voice modeling.

# From core/audio_generator.py class AudioNarrator: def synthesize_speech(self, text: str, voice_id: str = DEFAULT_VOICE_ID) -> Dict[str, Any]: # Deterministic MD5 hashing ensures instant offline replay from audio_cache/  text_hash = hashlib.md5(f"{voice_id}:{text}".encode("utf-8")).hexdigest() filename = f"stargaze_{text_hash[:12]}.mp3" ... 

Offline Caching & Harmonic Synthesis:

• Deterministic Disk Caching: Audio chunks are hashed and stored locally in audio_cache/. Once generated, the tour runs with zero network bandwidth.
• Procedural 432 Hz Drone: Native Web Audio API dual-oscillator synthesizer generates a continuous 432 Hz Pythagorean harmonic drone and solar wind filter beneath the spoken track.
• 528 Hz Solfeggio Chime: Spoken questions are acknowledged by a gentle 528 Hz sine wave bell with exponential gain decay, giving clear acoustic feedback without requiring visual cues.

4. Celestial Mechanics & 24-Hour Solar Phase Tracking (Astropy)

Instead of relying on pre-rendered planetarium approximations, WOTS runs high-precision astrodynamics locally using Astropy v7. It transforms International Celestial Reference System (ICRS) Right Ascension and Declination coordinates into local topocentric Altitude and Azimuth horizontal frames in real time:

# From core/celestial_engine.py from astropy.coordinates import EarthLocation, AltAz, SkyCoord, get_sun from astropy.time import Time import astropy.units as u def compute_local_sky(lat: float, lon: float, ra_hours: float, dec_deg: float): location = EarthLocation(lat=lat * u.deg, lon=lon * u.deg) altaz_frame = AltAz(obstime=Time.now(), location=location) # Real-time horizontal coordinate transformation  target = SkyCoord(ra=ra_hours * u.hourangle, dec=dec_deg * u.deg) local_target = target.transform_to(altaz_frame) return local_target.alt.deg, local_target.az.deg 

Real-Time 24-Hour Solar Phase Engine

By calculating the Sun's continuous ephemeris via get_sun(Time.now()), WOTS dynamically transitions its experience across all 24 hours of the day:

🌍 Why Does Open Innovation Matter?

This project proves an uncompromising axiom of outdoor computing: the wild has no cell towers, and nature has no API keys.

If outdoor technology requires an active internet connection to function, it is not outdoor technology—it is a tethered indoor application pretending to be adventurous.

1. Closed Cloud APIs Fail in the Wilderness: When an observer stands under the dark skies of Death Valley or the Scottish Highlands, proprietary cloud endpoints return connection timeouts. Open-weight models like Google Gemma and local foundation models like TabPFN run directly on the silicon in your backpack.
2. Absolute Privacy of Physical Sanctuary: Stargazing is an intimate, contemplative act. Observers should never be forced to transmit their real-time GPS coordinates, biometric sleep patterns, or nighttime habits to corporate telemetry servers. WOTS calculates all celestial and sensor mechanics locally.
3. Zero Token Anxiety: An observer can lie on the grass listening to the cosmos for eight continuous hours without paying per-token inference fees.

🏕️ Taking It Outside: Field Test

(Bonus points in action!)

On an autumn evening, I brought a laptop and smartphone setup out into the field at Mount Tamalpais Dark Sky Point (Bortle Class 3):

• Ambient Temperature: 10.5°C
• TabPFN Forecast: 93% atmospheric transparency, 0.82" FWHM seeing
• Observed Targets: Cassiopeia, Andromeda Galaxy (M31), Saturn, Cygnus, Summer Triangle

 ┌──────────────────────────────────────────────┐ │ FIELD OBSERVATION LOG ENTRY │ │ │ │ Location: Mount Tamalpais (Bortle 3) │ │ Time: 21:15 PST • Autumn Equinox Sky │ │ Hardware: Phone face-down on wool blanket │ │ Audio: Wired earbuds • George (Warm Voice) │ │ Screen Exposure: 0 seconds (0 Nits) │ │ │ │ "Andromeda M31 visible as an elongated mist │ │ to naked eyes after 22 minutes adaptation. │ │ Zero screen glare was used all evening." │ └──────────────────────────────────────────────┘ 

The Experience

I laid a wool blanket across the cold grass, plugged in earbuds, tapped "Start Audio Tour", and rested the phone face-down upon my chest.

The screen instantly engaged 0-nit true OLED darkness. With my eyes closed, the gentle 432 Hz harmonic drone settled my breathing. George's voice whispered softly:

"Welcome to your personal sky observatory... Your head is resting back toward the North. Slowly open your eyes straight up into the darkness..."

Because no screen light bleached my retinas, my eyes reached full scotopic adaptation after 20 minutes. Following the audio star-hop down from Cassiopeia's reclining 'W', I spotted the faint oval smudge of the Andromeda Galaxy using averted vision. When I wanted to check Saturn's rings, I simply asked aloud: "Where is Saturn?" A soothing 528 Hz ping sounded, and George whispered: "In tonight's sky, look toward the south-east at 22 degrees altitude..."

🤖 My Agent Session

This application was engineered collaboratively with an autonomous AI pair programming agent across multiple architectural sprints.

You can inspect the entire transcript, tool invocations, and commit trajectory:

• Session ID: 61027c66-9044-4d50-8a12-a251d0fe2c5c
• Agent Capabilities Used: Autonomous background execution, AST syntax validation, Git branch management, and real-time Python/Node verification.

🏆 Prize Categories

I am officially submitting Whispers of the Sky (WOTS) for the following partner prize categories:

1. 🥇 Best Use of Gemma

• Deployed Google's open-weight gemma2:2b via local Ollama inference (core/gemma_astronomer.py).
• Provides 100% offline edge astronomical reasoning, conversational star-hop generation, and live hands-free voice Q&A in wilderness environments with zero internet access.

2. 🥇 Best Use of TabPFN

• Integrated Prior Labs' tabular foundation models (TabPFNRegressor & TabPFNClassifier) in core/tabpfn_predictor.py.
• Ingests complex micro-meteorological atmospheric tables to predict Astronomical Seeing Quality (FWHM arcseconds), Naked-Eye Limiting Magnitude, and Sky Transparency Index (%), and models the 8-hour Golden Window timeline.

3. 🥇 Best Use of ElevenLabs

• Implemented ElevenLabs Neural Text-to-Speech (core/audio_generator.py) deploying George and Sarah models.
• Features low-timbre whispered audio calibration specifically tailored for midnight outdoor environments, backed by deterministic disk caching (audio_cache/) for trail playback.

4. ☁️ Render Cloud Deployment

• Complete 1-click cloud orchestration via render.yaml Blueprint.
• Automatically provisions Python 3.12, installs Astropy/FastAPI dependencies, and serves the live application at https://whispers-of-the-sky.onrender.com/.

5. 🌿 Week 1 Grand Prize: Touch Grass

• Embodies the core philosophy of the challenge: less than 30 seconds of screen time total.
• Encourages observers to lie down on living grass, breathe fresh night air, and look directly at ancient starlight with dark-adapted bare eyes.