v2.0.0 — Intelligent Vision

Drishtik v2.0.0 — Intelligent Vision

Highlights

• Local Vision AI — Ollama qwen2.5vl:7b on-device analysis, no cloud dependency
• 3-Pillar Alert System — Structured AI JSON + Rule-based severity scoring + Professional WhatsApp formatter
• Professional WhatsApp Alerts — Emoji severity, organized sections, snapshot + video clip attached
• Descriptive Alexa Announcements — Full security briefings instead of generic alerts
• Structured HA Notifications — Severity emoji, importance levels, rich MQTT payload
• Smart Media Decisions — Snapshot/clip/monitoring automatically selected by risk level
• Drishtik Control Panel — Web UI for all settings, diagnostics, reports, and logs
• Event Memory — JSONL history for pattern awareness across detections
• Multi-Step Reasoning — Re-confirms high/critical alerts with second AI pass
• Rule-Based Scoring — Deterministic severity adjustment using time, zone, home mode, behavior

Alert Filtering

• WhatsApp: medium/high/critical only
• Alexa: medium/high/critical only (daytime 6AM-11PM)
• HA Dashboard: all events with severity indicators

What Is New Since v1.0

• Local Ollama VLM replaces cloud-only GPT-4o-mini (with automatic fallback)
• 4-strategy JSON parser for robust AI output handling
• Severity scoring engine with contextual rules
• Professional formatted WhatsApp with snapshot + clip attachment
• Full web control panel with 12 management screens
• Camera context descriptions and zone-based policy
• Phase system (3-8) with independent toggles
• Runtime config file for all settings (no code editing needed)