# 🚀 VibePost Concierge: A Multi-Stage, Privacy-First Content Agent

VibePost Concierge is a local multi-stage content curation and safety auditing pipeline built on the **Google ADK Framework** and powered by **Gemini 3.5 Flash**. It solves the problem of manually curating media files for social media post by systematically scanning local folder, identifying the structural and visual context of engineering artifacts vs. personal media, executing pixel-level privacy guardrails, and compiling publication-ready bundles along with captions. This automation with the help of VibePost agent helps the users to curate and prepare social media ready drafts from bulk of media files which saves lot of time.

---

## 🧠 System Architecture & Workflow Flow

The script processes media inputs through four decoupled architectural stages:

1. **Multimodal Feature Extraction (Tool Node):** Programmatically loops through local directories and handles pixel-matrix inspections via Gemini Vision to generate structural context descriptors.

2. **Local Disk Cache Engine Persistence:** Intercepts out-of-box pipeline flows to check against a local serialization file (.gallery\_cache.json). New assets require exactly 1 cloud transaction; existing assets hit cache immediately, dropping network traffic to zero and mitigating 429 RESOURCE\_EXHAUSTED caps.

3. **Deep Privacy Security Guardrail (Circuit Breaker):** Actively reviews visual extraction streams for programmatic configuration exposures (AWS secrets, API tokens, configuration credentials). Detected leaks trip an immediate operational circuit breaker, safely killing the thread execution.

4. **Contextual Curation & Formatting Engine (Core Agent):** Isolates personal lifestyle artifacts from system engineering snapshots based on the chosen platform stream selection (Instagram Lifestyle vs. LinkedIn Engineering).

---

## 🛠️ Local Installation & Usage

### Step 1: Project Initialization

Open your terminal inside the root folder and run:

\# Initialize and activate Python environment sandbox

python -m venv venv

.\\venv\\Scripts\\Activate.ps1



\# Install core required framework packages

pip install google-adk pillow python-dotenv

### Step 2: Configure Environment Variables

Create a hidden file named .env inside your vibe\_post\_agent/ directory:

GOOGLE\_API\_KEY="YOUR\_ACTUAL\_GEMINI\_API\_KEY\_HERE"

### Step 3: Execution Loop

Drop your raw gallery images or technical screenshots into the automatically tracked ./input\_gallery/ folder, then trigger the central orchestrator:

python run\_pipeline.py

---

## 🛡️ Hardened Edge-Case Testing Records

* **Scenario 1 (Security Breached):** Verified via explicit testing with hardcoded text token configurations in screenshot files. The agent catches visual strings, flips the internal schema payload to "is\_safe": false, and aborts the pipeline safely.

* **Network Tolerance Retries:** Implements an adaptive exception handler loop that traps multi-threaded network surges, pausing gracefully rather than dropping execution threads.