SocialBytes Intelligence Agency 🚀

Powered by crewai orchestrating an agentic AI team or research agents

Automated media intelligence for the post-truth era.
SocialBytes is an agentic AI workspace that autonomously researches, analyzes, and reformats complex web content into actionable intelligence briefs. By leveraging multi-vector sentiment analysis and semantic drift checks, it helps agencies and analysts see through the noise.

🏗 System Architecture

The SocialBytes Engine uses a hybrid architecture, combining a 2-Agent AI Crew with deterministic system processors:

Phase 1: The Crew (Agentic AI)

Powered by crewai & GPT-4o

1. The Inquisitor (Research Agent): Performs neural searches (Exa) to find related articles, critiques, and "hidden facts" that the primary source omits. It maps the Narrative Horizon by identifying aligned vs. counter-narrative sources.
2. The Stylist (Persona Agent): Transforms dry research data into high-converting social content (LinkedIn, X, Instagram). It dynamically switches personas (Advocate vs. Contrarian) based on the user's strategic intent.

Phase 2: The Infrastructure (System Agents)

Deterministic Algorithms & Proxies

1. The Curator (Media Proxy): A custom asset manager that scrapes, dedupes, and bundles high-resolution media assets into a secure gallery, bypassing hotlink protections.
2. The Publisher (Export Engine): A rendering engine that generates enterprise-grade PDF briefs with automated APA-lite citations and comparative sentiment tables.

All components communicate via a shared Task Context Chain, ensuring that insights found by the AI agents are structurally preserved for the export engines.

⚡ The Tech Stack

Frontend (Chrome Extension & Web Dashboard)

• React + Vite: Ultra-fast rendering with optimized state management.
• Tailwind CSS: Custom "Deep Indigo / Tech-Orange" theme for a premium, futuristic verified aesthetic.
• Lucide Icons: Consistent, lightweight visual language.
• Chrome Manifest V3: Deep-link auto-pilot allows users to trigger research directly from their browser toolbar.

Backend (Intelligence Engine)

• Python (FastAPI): High-performance async API server.
• CrewAI: Multi-agent orchestration framework.
• Neural Search (Exa/OpenAI): Semantic search capabilities to find thematically similar content rather than just keyword matches.
• JSZip & Blob Proxy: Client-side bundling of media assets to ensure 403-safe downloads without server-side storage.

Platform Infrastructure

• Docker: Containerized service mesh for Engine and Dashboard.
• Browser Security: Implemented CORS proxies and Blob-URL manufacturing to bypass scraping protections securely.

🌟 Core Functionalities

1. Multi-Vector Sentiment Analysis

Unlike simple positive/negative classifiers, SocialBytes maps content against a Narrative Horizon. It identifies:

• Perspective Gap: The divergence between the article's claims and the broader market consensus.
• Counter-Narratives: Automatically flags sources that dispute the primary article's premises (marked with Orange badges).

2. Media Intelligence Gallery

• Secure Deduplication: Uses a Set-based filter on image URLs to purge redundant thumbnails.
• High-Res Recovery: Fetches full-resolution assets via a proxy to bypass hotlinking protection.
• Bulk Export: JSZip integration allows one-click downloading of entire media kits.

3. Enterprise PDF Export

• Automated Citations: Generates an APA-lite reference list for every claim.
• Comparative Tables: Injects a structured analysis table (Source vs. Sentiment vs. Confidence) directly into the PDF.
• Print-Audit Safe: Using a hidden "Ghost Window" strategy, PDFs are rendered with zero UI artifacts (buttons, inputs) and perfectly paged layouts.

🧪 Test Validation & Assurance

We rely on a suite of validation scripts to ensure intelligence accuracy:

Script A: validate-api-latency.js

• Purpose: Measures the time-to-insight for the Research Crew.
• Metric: Ensures the "Deep Research" phase completes within the 45s SLA threshold.

Script B: check-data-integrity.py

• Purpose: Semantic Drift Check.
• Method: Compares the cosine similarity of the generated brief against the source text to ensure the AI hasn't hallucinated facts.

Script C: media-dedup-test.sh

• Purpose: Asset Integrity.
• Method: Simulates a scrape of a media-heavy page and verifies that the final ZIP bundle contains only unique file hashes.

🗺 Future Roadmap

• Semantic Caching (Redis): To reduce API costs and latency for frequently researched topics.
• Agency Expansion:
  • FinTech: "Due Diligence" Agent for analyzing earnings reports.
  • HealthTech: "MedVantage" Agent for parsing clinical trials.
• Deep-Link Sharing: Generate shareable, read-only links to research briefs for team collaboration.

🧠 Just-in-Time RAG Architecture

SocialBytes eschews traditional vector databases for a Stateless Neural Retrieval model suitable for breaking news.

• The Problem: Traditional RAG (Retrieval-Augmented Generation) relies on pre-indexed benchmarks which become stale instantly for news.
• The Solution: The Inquisitor agent performs Real-Time Retrieval Augmentation.
  1. Query Generation: The LLM decomposes the user's URL into search intent vectors.
  2. Live Extraction: Exa's Neural Search provides live text chunks from the last 24-72 hours.
  3. Context Injection: These chunks are injected into the 'Stylist' agent's context window, eliminating hallucinations and ensuring the brief reflects the current state of the world.

🚀 Installation & Quick Start

Get the agency up and running in 4 steps:

1. Clone the Repository

   ```bash
   git clone https://github.com/5ync/socialbytes-agency.git
   cd socialbytes-agency
   

2. Install Dependencies

   • Backend: cd api-engine && pip install -r requirements.txt
   • Frontend: cd web-dashboard && npm install

3. Set Environment Variables Create a .env file in the root:

   OPENAI_API_KEY=sk-...
   EXA_API_KEY=...
   MODEL_NAME=gpt-4o
   

4. Launch the Suite

   • Docker (Recommended): docker-compose up --build
   • Manual: uvicorn src.main:app --reload (port 8000) and npm run dev (port 5173).

🔧 Troubleshooting & Resiliency

• Browser Extension:
  • Issue: "ActiveTab" permission not firing.
  • Fix: Ensure the extension is pinned. Reload the extension in chrome://extensions if the Service Worker goes dormant.
• Media Assets (403 Forbidden):
  • Strategy: SocialBytes uses a Blob Proxy. Instead of hotlinking, the backend fetches the image, converting it to a Base64 stream, which the frontend wraps in a Blob URL. This creates a local, secure reference that bypasses CDN headers.
• PDF Rendering:
  • Strategy: To avoid cutting off text or including UI buttons, we use a Ghost Window. This invisible iframe renders a print-specific DOM off-screen, waits for all assets to load via Promise.all(), and then triggers the print dialog.

🔒 License

This project is proprietary. Please see the LICENSE file for usage restrictions.

Built with ❤️ by Jorge Barrera
Engineered as a high-fidelity intelligence ecosystem for the 2026 capstone, focusing on agentic orchestration and narrative integrity.