Project: Multi Agent Content Generation

[bhanuxbisht]

Track

Creative Apps (GitHub Copilot)

Project Name

(bb /create) Multi Agent Content Generation

GitHub Username

bhanuxbisht

Repository URL

https://github.com/bhanuxbisht/agent

Project Description

bb /create is a multi-agent AI content production engine that turns a single
topic prompt into a full production-ready content blueprint — script, shot
timeline, visual enhancements, story structure, and critique score.

Problem it solves: Content creators waste hours manually scripting, planning
timelines, and structuring short-form video. bb /create automates the entire
pre-production pipeline in seconds.

Key features:

• 7 specialized LangGraph agents: Analyzer → Script Writer → Timeline Planner
  → Enhancement → Story Architect → Critic → Refiner
• Self-improving quality loop: Critic scores output 1-10; Refiner rewrites
  weak sections until score ≥ 7 (max 3 iterations)
• Real-time SSE streaming so users watch each agent work live
• Supports Instagram, YouTube, TikTok, and general platforms
• Content types: reel, short, YouTube video, film, podcast
• FastAPI REST + streaming backend; React + Tailwind frontend
• Fully containerised with Docker Compose

Demo Video or Screenshots

Demo Video : https://youtu.be/wYcmemBU1EM?si=WjtODmE7INV1tkY2
Screenshots: https://github.com/bhanuxbisht/agent/tree/main/screenshots

Primary Programming Language

Python

Key Technologies Used

• LangGraph (Agent Orchestration)
• LangChain
• Groq Cloud API (llama-3.3-70b-versatile)
• FastAPI (Backend with SSE streaming)
• React.js / Vite (Frontend)
• Tailwind CSS
• Docker & Docker Compose

Submission Type

Individual

Team Members

No response

Submission Requirements

• My project meets the track-specific challenge requirements
• My repository includes a comprehensive README.md with setup instructions
• My code does not contain hardcoded API keys or secrets
• I have included demo materials (video or screenshots)
• My project is my own work with proper attribution for any third-party code
• I agree to the Code of Conduct
• I have read and agree to the Disclaimer
• My submission does NOT contain any confidential, proprietary, or sensitive information
• I confirm I have the rights to submit this content and grant the necessary licenses

Quick Setup Summary

1. Clone the repository
2. Navigate to the project root and duplicate .env.example to .env
3. Add your Groq API Key to the .env file
4. Start the application using Docker: docker compose up --build
5. Open http://localhost in your browser

Technical Highlights

• Built a multi-agent orchestrated pipeline using LangGraph, executing 7 explicit nodes (Analyzer -> Writer -> Planner -> Enhancer -> Architect -> Critic -> Refiner).
• Developed an automated quality-control loop where a "Critic" agent scores the output and reroutes back to a "Refiner" agent if the score is under 7/10.
• Implemented Server-Sent Events (SSE) to stream the LangGraph execution steps directly to a React frontend, allowing the user to watch the AI agents deliberate and construct the production blueprint in real-time.
• Forced plain-text JSON parsing over LLM Tool Function calling to drastically improve speed and avoid API schema-failure errors with Llama 3 on Groq.

Challenges & Learnings

Biggest Challenge: Handling structured output reliability from the open-source model. The model would frequently wrap valid JSON inside markdown blocks or XML function tags, causing the strict parser to crash.

Solution: I had to implement a custom robust _safe_json_parse middleware that detects and strips markdown fences, catches decoding errors, and falls back to safe schemas instead of failing the whole pipeline. I also migrated away from strict with_structured_output() functions and prompted the LLM natively for JSON, which lowered the failure rate to zero.

Key Learning: LangGraph's state routing is incredibly powerful for establishing feedback loops. Letting one agent critique another agent before showing the user the final output drastically improves the quality of the generated content.

Contact Information

workforrhody7@gmail.com

Country/Region

India