AI Analysis

Overview

AI analysis is Evatar's core intelligence capability. Every uploaded screenshot automatically enters the analysis pipeline, where LLM performs multimodal analysis to extract structured information including app name, content category, user intent, summary, entities, and confidence.

Analysis Fields

Field	Type	Description	Example
app_name	String	Screenshot source app name	"WeChat", "Alipay", "12306"
content_category	String	Content category	"chat", "webpage", "finance", "notification"
intent	String	User intent	"reminder", "research", "reference", "note", "ignore"
relevance	String	Relevance level	"high", "medium", "low"
summary	String	Summary (2-3 sentences)	"User discussed tomorrow's meeting with Zhang San in WeChat"
entities	Array	Extracted entity list	[{"type":"name","value":"Zhang San"}]
confidence	Float	Confidence (0.0-1.0)	0.85

Analysis Pipeline Flow

1. Screenshot upload completes
2. enqueue_analysis creates async task
3. _safe_process with 3 retries (exponential backoff)
4. process_photo() checks idempotency, encodes image, calls LLM
5. Parses JSON response, saves Analysis record
6. If relevance is not low and confidence >= 0.3, extracts memory
7. Every 3 completed analyses triggers intent reasoning

Analysis Status Flow

Status	Description
pending	Waiting for analysis
processing	Currently analyzing
done	Analysis complete
error	Analysis failed (retryable)