ET Multiple Agent

ET Multiple Agent is a multi-step AI content pipeline for generating blogs, reviews, social posts, images, scheduling metadata, and post-approval translations from a single web UI.

The project uses:

• FastAPI for the backend API
• LangGraph for the generation workflow
• Groq for LLM-powered writing, review, and translation
• Tavily and NewsAPI for news-mode research
• Bytez as the preferred image generation provider
• Stability AI as an image-generation fallback
• a small C++ localization engine wrapper plus Groq fallback translation

Features

• Generate content in two modes:
  • news
  • product
• Multi-step content pipeline:
  • write
  • validate
  • RAG validation
  • review
  • image generation
  • social post generation
  • human approval loop
  • localization
• Human-in-the-loop approval and refinement
• Translation support after approval
• Social post scheduling endpoints
• Single-page frontend with tabs for:
  • blog
  • review
  • social posts
  • images
  • translations
  • schedule queue

Project Structure

ETHACKATHON/
|-- agents/                    # Pipeline agents
|-- engine/                    # C++ localization engine source
|-- graph/                     # LangGraph workflow
|-- prompts/                   # Prompt templates
|-- rag/                       # RAG helpers and vector store integration
|-- api_server.py              # FastAPI backend
|-- index.html                 # Frontend UI
|-- config.py                  # Environment-backed config
|-- start.ps1                  # Windows startup script
|-- run.bat                    # Windows launcher
`-- test_graph_state.py        # Direct localization test

Requirements

• Windows
• Python 3.12+ recommended
• A virtual environment at ETHACKATHON\.venv

API keys are read from .env.

Environment Variables

Create a .env file in the project root with:

GROQ_API_KEY=your_groq_key
GROQ_MODEL=llama-3.3-70b-versatile
TAVILY_API_KEY=your_tavily_key
STABILITY_API_KEY=your_stability_key
BYTEZ_API_KEY=your_bytez_key
BYTEZ_IMAGE_MODEL=stabilityai/stable-diffusion-xl-base-1.0
NEWSAPI_KEY=your_newsapi_key
HF_TOKEN=optional_huggingface_token

Notes:

• GROQ_API_KEY is required for blog generation, review, and translation.
• GROQ_MODEL is optional. If omitted, the app falls back to the default from config.py.
• TAVILY_API_KEY is used for news research.
• NEWSAPI_KEY supports additional news retrieval.
• BYTEZ_API_KEY is the preferred key for image generation.
• BYTEZ_IMAGE_MODEL is optional and defaults to stabilityai/stable-diffusion-xl-base-1.0.
• STABILITY_API_KEY is optional fallback support for image generation.
• HF_TOKEN is optional in the current codebase.

Installation

From the project root:

python -m venv .venv
.\.venv\Scripts\activate
pip install -r requirements.txt

Running the Project

Option 1: Recommended

Use the included launcher:

.\start.ps1

or

run.bat

This starts:

• backend at http://127.0.0.1:8000
• frontend at http://127.0.0.1:5500/index.html

start.ps1 now clears stale listeners on ports 8000 and 5500 before launching, which helps avoid old backend/frontend instances hanging around between restarts.

Option 2: Manual

Start backend:

.\.venv\Scripts\python.exe -m uvicorn api_server:app --host 127.0.0.1 --port 8000

Start frontend:

.\.venv\Scripts\python.exe -m http.server 5500

Then open:

http://127.0.0.1:5500/index.html

Workflow Overview

The backend pipeline is defined in graph/blog_graph.py.

Main flow:

1. write
2. validate
3. rag_validate
4. review
5. gen_images
6. gen_social
7. human_review
8. localize after approval

Important behavior:

• Translation only runs after the user approves the content.
• The frontend sends selected target languages during generate and approval.
• Social image generation reuses already-generated platform images when available instead of regenerating them.
• Image generation runs selected formats in parallel.
• Localization first pings the C++ layer if present, then performs the actual translations through Groq in parallel.

API Endpoints

POST /api/generate

Starts a new generation job.

Example fields:

• mode
• topic
• audience
• length
• context
• product_details
• key_features
• uvp
• generate_images
• image_formats
• social_platforms
• user_image_b64
• target_languages

GET /api/status/{job_id}

Returns the current job state, including:

• generation status
• parsed blog
• review details
• image outputs
• social posts
• localized_content

POST /api/feedback

Used for:

• approve
• refine

When approved, localization is triggered if target_languages are present.

POST /api/schedule

Schedules a generated social post.

GET /api/queue

Returns the current scheduled post queue.

GET /api/health

Simple health endpoint.

Translation Notes

Translation support is implemented through:

• graph/blog_graph.py
• agents/localization_wrapper.py
• engine/localization_agent.cpp

Behavior:

• if no target languages are selected, localization is skipped
• if the content is not approved, localization is skipped
• after approval, translated content is stored under localized_content

Testing

You can directly test the localization path with:

.\.venv\Scripts\python.exe test_graph_state.py

This exercises run_localization(...) directly.

Common Issues

Translation not working

Check:

• .env exists in the project root
• GROQ_API_KEY is valid
• target languages are selected in the UI
• content is approved, not just generated

Notes:

• if engine/localization_agent.exe is missing, the app now falls back to Groq-only translation
• translations are generated after approval, not during the first draft pass

News mode has weak context

Check:

• TAVILY_API_KEY
• NEWSAPI_KEY

Images not generated

Check:

• BYTEZ_API_KEY
• STABILITY_API_KEY
• image generation toggle in the UI

Notes:

• Bytez is the preferred provider
• Stability is only used as fallback
• if Instagram/LinkedIn images were already generated in the image stage, the social stage reuses them for speed

App starts but frontend cannot connect

Check:

• backend is running on 8000
• frontend is running on 5500
• browser is using http://127.0.0.1:5500/index.html

Notes For GitHub

Generated runtime artifacts are intentionally ignored:

• .env
• __pycache__
• logs
• generated outputs
• local Chroma DB files
• compiled localization executable

This keeps the repository focused on source code.