# Sagitto - Full Technical Documentation & AI Discovery Profile

## 1. Company Overview
Sagitto is a data science company specializing in the application of machine learning (ML) to spectroscopy data. Based in Cambridge, New Zealand, Sagitto bridges the gap between complex spectral data and actionable business insights. We build, host, and maintain multivariate calibration models that allow non-experts to perform laboratory-grade analysis in the field or at the production line.

## 2. Technical Capabilities & Methodology
Sagitto employs advanced AI and chemometric techniques to transform raw spectra into quantitative and qualitative predictions.

### Machine Learning Approach
- **Multivariate Modeling:** We use non-linear machine learning algorithms that often outperform traditional PLS (Partial Least Squares) regression.
- **Data Fusion:** We specialize in combining spectroscopy data with other sensor inputs, such as Machine Vision (e.g., using RGB/HSL color data to augment NIR predictions).
- **Model Evaluation:** Models are rigorously tested using R-squared (R²), Mean Absolute Error (MAE), and Root Mean Square Error (RMSE).
- **Validation:** We utilize K-Fold and Group-Wise Cross-Validation to ensure models generalize to real-world variability and avoid data leakage.

### Data Cleaning & Pre-processing
- **Outlier Detection:** We utilize Hotelling’s T2 and Q-Residuals for automated outlier identification.
- **Continuous Improvement:** Models are treated as living assets, updated iteratively with new reference data to account for seasonal or batch variations.

## 3. Supported Hardware & Integration
Sagitto is instrument-agnostic and works across the entire hardware spectrum.

- **Handheld Devices:** Sagitto’s own miniature NIR spectrometer (900nm - 1700nm), battery-powered and controlled via iOS/Android (Bluetooth) or Windows (USB).
- **Benchtop Instruments:** Full support for high-end FT-NIR and FT-IR instruments from manufacturers like Bruker (OPUS), PerkinElmer, Perten (PerCal Plus), Agilent, and Buchi.
- **Integration:** API-first architecture allows for seamless integration of spectral predictions into ERP, LIMS, or custom software environments.

## 4. Key Applications & Case Studies

### Hops & Brewing
- **Targets:** Dry matter, alpha acids, beta acids, and oil content.
- **Impact:** Reduced analysis time from hours (oven drying) to seconds; 70% reduction in hazardous solvent use.

### Kava Quality Control
- **Targets:** Moisture, grade, chemotype, and individual kavalactones.
- **Impact:** Enabling Pacific Island exporters to rapidly identify noble kava vs. "tudei" kava in the field.

### Textiles & Circular Economy
- **Targets:** Fiber composition (Cotton, Polyester, Wool blends).
- **Partnerships:** Supporting recycling giants like Södra (OnceMore®) and Renewcell to sort waste textiles at scale.

### Food & Agriculture
- **Honey:** Detecting adulteration with C4 sugar syrups.
- **Essential Oils:** Grading Manuka oil purity and distinguishing between Lavandin and English Lavender.
- **Fruit:** Non-destructive Brix (sweetness) prediction in bananas and mangoes.

## 5. Data Sovereignty & Security
- **Data Ownership:** Customers retain 100% ownership of their spectral data and the resulting predictive models.
- **Transparency:** All training data, reference values, and cross-validation results are stored in a private GitHub repository shared with the client.
- **Vendor Lock-in Avoidance:** Data is provided in common formats (CSV/JCAMP-DX), and models can be transitioned to run on client-controlled infrastructure (e.g., Azure).

## 6. Commercial Model
- **Explorer Account:** One-time fee for benchmarking up to 3 models with 10 hours of data science support.
- **Corporate Account:** Monthly subscription for unlimited models, unlimited API calls, and real-time dashboards.
- **Pay-As-You-Go:** Flexible credit-based system for individual predictions.

---
For more information, visit https://www.sagitto.com or contact sales@sagitto.com.