LogitGuard-The AI workflow Guardian

Inspiration

The idea came from a desire to build a hybrid architecture. I wanted to try something different.

How We Built It

LogiGuard was built as a hybrid AI Studio + Antigravity + Cloud Run system designed to treat screenshots and screen recordings as analyzable data streams. We started by designing a clean, event‑driven architecture:

• A lightweight backend that receives images or videos and forwards them to Gemini Multimodal.
  • A reasoning layer that interprets UI states, risk signals, and workflow events.
  • A normalization layer that converts raw model output into a consistent event schema.
  • A production-grade Antigravity agent that enforces strict JSON output and tool‑aware behavior.
  • A set of public endpoints (/health, /status, /analyze-visual, /events) that make the system easy to evaluate, demo, and integrate. We iterated quickly, building the system like Lego bricks:
    First the backend, then the prompts, then the event schema, then the agent, then the demo flow, and finally the submission assets.

## LogiGuard is a hybrid system combining an AI Studio, Antigravity, and Cloud Run, specifically engineered to process screenshots and screen recordings as analyzable data streams. Our foundational design is clean, event-driven architecture, structured around these key components: Lightweight Backend: Responsible for receiving visual data (images or videos) and forwarding it to Gemini Multimodal. Reasoning Layer: Interprets complex inputs such as UI states, potential risk signals, and workflow events. Normalization Layer: Standardizes the raw output from the model into a consistent event schema. Production-Grade Antigravity Agent: Enforces strict JSON output compliance and ensures tool-aware behavior. Public Endpoints: A set of accessible endpoints (/health, /status, /analyze-visual, /events) designed for easy testing, demonstration, an integration. We adopted a rapid, modular development approach, building the system like Lego blocks in this sequence: the backend, the prompts, the event schema, the agent, the demonstration flow, and finally, the submission assets.

Accomplishments We Are Proud Of

1. A fully working visual analysis pipeline From screenshot → Gemini → normalized events → risks → logs. It works end‑to‑end and feels surprisingly natural.
2. A production-grade Antigravity agent The agent is strict, predictable, and ready for real workflows — not just a demo.
3. A clean, API-first architecture Judges can test LogiGuard with simple curl commands and immediately see value.
4. A compelling, judge-ready demo The phishing scenario shows exactly why visual analysis matters.
5. A unified submission bundle

What We Learned

1. Screens are a goldmine of signals UI states reveal security risks, compliance issues, and product failures long before logs do.
2. Multimodal reasoning becomes powerful when you enforce structure Raw model output is interesting. Structured model output is useful.

3. Prompt engineering is software engineering We treated prompts like code: versioned, assessed, refined, and validated.

   1. Hybrid architecture is the future AI Studio for reasoning design. Antigravity for production execution. Cloud Run for scalable hosting. Together, they create a system that feels modern and robust.
   2. The best demos tell a story Judges do not remember features — they remember moments. LogiGuard creates those moments.

## What’s Next for LogiGuard LogiGuard v1 is just the beginning. Here is where we are taking it next:

1. Real-time session monitoring Continuous visual analysis of a user’s screen or workflow.
2. Custom rule packs Security teams, compliance teams, and QA teams can define their own risk rules.
3. Dashboard for analysts A visual timeline of events, risks, and UI states — like a flight recorder for screens.
4. Integration with SIEM and alerting tools Send high-severity risks directly into Slack, PagerDuty, or Splunk.
5. Long-form video understanding Analyze full sessions, not just short clips.
6. Organization-level memory Learn patterns across teams, apps, and environments.

Built With

• ai-studio
• antigravity
• cloud-run
• javascript
• json
• typescript
• yaml