Skip to content

DEVELOPMENT_PLAN.md

Latest commit

 

History

History
289 lines (225 loc) · 7.94 KB

DEVELOPMENT_PLAN.md

File metadata and controls

289 lines (225 loc) · 7.94 KB

LARUN Development Plan

╔══════════════════════════════════════════════════════════════════════════╗
║  LARUN TinyML - Development Roadmap                                      ║
║  Created by: Padmanaban Veeraragavalu (Larun Engineering)               ║
╚══════════════════════════════════════════════════════════════════════════╝

Current Status (v2.0.0)

Completed

  • Core CLI with skills system
  • Chat interface (larun_chat.py)
  • NASA data fetching (TESS/Kepler via lightkurve)
  • Basic transit detection model (81.8% accuracy)
  • TFLite export for edge deployment
  • Distributed training system
  • Google Colab notebook for GPU training
  • Code generation addon

Models Available

Model Size Accuracy Location
Transit Detector 50 KB 81.8% models/real/astro_tinyml_real.tflite
Quantized INT8 ~25 KB ~80% models/real/astro_tinyml_real_int8.tflite

Phase 1: Core Algorithm Improvements (Priority: HIGH)

1.1 BLS Periodogram Implementation

Reference: docs/research/EXOPLANET_DETECTION.md

Status: COMPLETE
Complexity: Medium
Timeline: Sprint 1

Tasks:

  • Implement Box Least Squares algorithm in src/skills/periodogram.py
  • Add period grid optimization for efficiency
  • Integrate with astropy.timeseries.BoxLeastSquares
  • Create CLI command: larun analyze bls (also /bls)
  • Add unit tests with synthetic transit data
  • Benchmark against published Kepler results

Expected Outcome:

  • Detect periodic transits with SNR > 7
  • Support period range 0.5 - 100 days
  • False alarm probability calculation

1.2 Phase Folding & Transit Fitting

Reference: docs/research/EXOPLANET_DETECTION.md

Status: COMPLETE
Complexity: Medium
Timeline: Sprint 1-2

Tasks:

  • Implement phase folding algorithm (src/skills/periodogram.py)
  • Add binning for visualization
  • Integrate batman for transit model fitting (src/skills/transit_fit.py)
  • Fit: Rp/Rs, a/Rs, inclination, limb darkening
  • Create CLI command: /fit, /phase

1.3 False Positive Identification

Reference: docs/research/EXOPLANET_DETECTION.md

Status: COMPLETE
Complexity: High
Timeline: Sprint 2

Tasks:

  • Odd-even transit depth test (src/skills/vetting.py)
  • Secondary eclipse search
  • Grazing binary detection (V-shaped)
  • Duration test
  • Create CLI command: /vet
  • Centroid shift analysis (requires TPF data)
  • Create FPP (False Positive Probability) calculator

Phase 2: Model Training Improvements (Priority: HIGH)

2.1 Increase Training Data

Current: ~150 samples (mixed real + synthetic)
Target: 500+ samples each class
Next retraining: After Phase 1 complete

Tasks:

  • Fetch 500+ confirmed exoplanets from NASA Archive
  • Implement data augmentation (src/augmentation.py)
  • Add K-fold cross-validation
  • Target: 90%+ validation accuracy

2.2 Model Architecture Optimization

Reference: docs/research/TINYML_OPTIMIZATION.md

Tasks:

  • Experiment with different architectures:
    • Deeper CNN (more layers, fewer filters)
    • 1D ResNet blocks
    • Attention mechanisms
  • Knowledge distillation from larger model
  • Pruning and quantization-aware training
  • Target: <50KB with 90% accuracy

2.3 Multi-class Classification

Current: Binary (planet/no-planet)
Target: 6-class classification

Classes:

  1. Confirmed Planet
  2. Eclipsing Binary
  3. Variable Star
  4. Stellar Activity
  5. Instrumental Artifact
  6. No Signal

Phase 3: New Data Sources (Priority: MEDIUM)

3.1 Gaia DR3 Integration

Reference: docs/integrations/GAIA_INTEGRATION.md (to create)

Tasks:

  • Add astroquery.gaia support
  • Fetch stellar parameters (Teff, logg, metallicity)
  • Cross-match with TIC catalog
  • Calculate stellar radii for planet radius estimation
  • Create skill: larun data gaia

3.2 JWST Data Access

Reference: docs/integrations/JWST_INTEGRATION.md (to create)

Tasks:

  • Access JWST spectra via MAST
  • Atmospheric transmission spectra parsing
  • Integration with exoplanet characterization
  • Create skill: larun data jwst

3.3 Ground-based Survey Integration

Tasks:

  • TESS Follow-up Observing Program (TFOP) data
  • ExoFOP integration
  • Radial velocity data access

Phase 4: Galaxy Classification (Priority: MEDIUM)

4.1 Galaxy Morphology CNN

Reference: docs/research/GALAXY_CLASSIFICATION.md

Tasks:

  • Download Galaxy Zoo dataset
  • Create image preprocessing pipeline
  • Train CNN for morphology classification:
    • Spiral
    • Elliptical
    • Irregular
    • Merger
  • Convert to TFLite (<100KB)
  • Create skill: larun classify galaxy

4.2 Galaxy Redshift Estimation

Tasks:

  • Photometric redshift from colors
  • Integration with SDSS data

Phase 5: Advanced Features (Priority: LOWER)

5.1 Multi-planet Detection

Reference: docs/research/EXOPLANET_DETECTION.md

Tasks:

  • Iterative transit removal
  • Transit Timing Variations (TTV) detection
  • Search for additional planets in residuals

5.2 Exomoon Search

Tasks:

  • Transit timing/duration variations
  • Photodynamic modeling
  • (Research-level complexity)

5.3 Spectral Analysis

Reference: docs/research/SPECTROSCOPY.md

Tasks:

  • Spectral line identification
  • Radial velocity extraction
  • Stellar classification from spectra

Phase 6: Deployment & Integration (Ongoing)

6.1 Edge Deployment

Reference: docs/research/TINYML_OPTIMIZATION.md

Targets:

  • Raspberry Pi deployment guide
  • ESP32 deployment (TensorFlow Lite Micro)
  • Mobile app (TFLite Android/iOS)

6.2 Web Dashboard

Tasks:

  • Interactive dashboard enhancement
  • Real-time data visualization
  • Cloud deployment option

6.3 API Service

Tasks:

  • FastAPI REST service
  • Docker containerization
  • Cloud deployment (GCP/AWS)

Sprint Schedule

Sprint Focus Duration
Sprint 1 BLS + Phase Folding 2 weeks
Sprint 2 False Positive + Model Training 2 weeks
Sprint 3 Gaia Integration + Multi-class 2 weeks
Sprint 4 Galaxy Classification 2 weeks
Sprint 5 Edge Deployment + API 2 weeks

Skill Implementation Checklist

For each new skill, follow docs/skills/SKILL_DEVELOPMENT.md:

  • YAML definition in skills/
  • Python implementation in src/skills/
  • Unit tests in tests/
  • CLI integration in larun.py
  • Documentation update
  • TinyML optimization check

Research Documentation Status

Document Status Priority
EXOPLANET_DETECTION.md Complete Reference
NASA_DATA_SOURCES.md Complete Reference
TINYML_OPTIMIZATION.md Complete Reference
SKILL_DEVELOPMENT.md Complete Reference
MAST_INTEGRATION.md Complete Reference
GALAXY_CLASSIFICATION.md Complete Reference
IMAGE_PROCESSING.md Complete Reference
STELLAR_PHYSICS.md Complete Reference
SPECTROSCOPY.md Complete Reference
GAIA_INTEGRATION.md To Create Sprint 3
JWST_INTEGRATION.md To Create Sprint 3

Next Immediate Actions

  1. Test current model - Run /detect with real target
  2. Start BLS implementation - High priority algorithm
  3. Improve training data - Use Colab for larger dataset
  4. Create Gaia integration docs - Needed for stellar parameters

Created by: Padmanaban Veeraragavalu (Larun Engineering) With AI assistance from Claude (Anthropic) Last Updated: January 2026