Architecture

FrogSense is written to make observations both low-friction and high-value.

Observation Sources

• Written text
• Voice recording (in app)
• Images

Observations can be captured directly or derived from evidence. For example, FrogSense can extract a configured measurement from an image while retaining the original image as evidence.

Signal Types

Observations do not need to be structured when they are captured. FrogSense preserves what was observed first.

Where useful, information can be represented as configured signals. Signals can define things such as expected values, units, validation rules, and instructions for extracting values from evidence.

This allows structure to be added where it provides value without requiring every observation to fit a predefined schema.

Provenance

FrogSense preserves the original evidence used to create an observation whenever available.

Evidence such as images and audio remains attached to the resulting observation rather than being discarded after extraction.

Additional attachments can be added to any observation.

Attachment storage is pluggable and currently supports local filesystem and S3-compatible object storage.

AI

• AI is optional; OpenAI is currently supported.
• AI can extract configured signals from evidence such as images.
• Extracted values pass through FrogSense’s normal validation before becoming observations.
• AI can also summarize longitudinal history or analyze observations for specific questions.
• Original evidence is retained independently of the AI-generated result.

In a nutshell

Evidence / Input → Observation → Optional Signal → Validation → Longitudinal History → Analysis