promoflow-ai

# PromoFlow AI 🎙️🌊  
**An AI-powered system that automates podcast promotion workflows using GPT-4 and structured prompt chains.  
Built with agent-first architecture to generate short-form clips, social captions, metadata, and thumbnail prompts from long-form content — reducing production time while preserving brand voice and emotional resonance.**

---

## 💡 Overview  
PromoFlow AI accelerates podcast content workflows by auto-generating ready-to-publish assets from long-form video/audio.  

This semi-automated system blends large language model (LLM) capabilities with human-in-the-loop QA to ensure quality, tone alignment, and rapid turnaround.

---

## 🧩 Problem → Solution → Results  

**Problem**  
Producing marketing assets from weekly podcast episodes was time-consuming and inconsistent — requiring multiple team members to manually extract highlights, write captions, and format deliverables.

**Solution**  
Designed a modular AI workflow that parses transcripts, identifies key moments, and generates clip selections, titles, captions, and thumbnail prompts in a consistent JSON format — all routed through a human QA checkpoint.

**Results**  
- Reduced asset production time by **60%**, saving **4–6 hours per episode**  
- Enabled faster go-to-market for weekly podcast releases  
- Improved handoff clarity between creative and marketing teams  
- Ensured consistent branding across platforms without additional headcount

---

## ✨ Key Features  
- Extracts **timestamps and themes** from podcast transcripts  
- Suggests **3–5 emotional hooks** for short-form content  
- Produces **titles, captions, and thumbnail prompts** tailored to social platform styles  
- Generates **Classic, Hybrid, and Modern** caption styles per client brand voice  
- Exports **JSON handoff format** for downstream automation  
- Supports **human-in-the-loop QA** to refine before publishing

---

## 🧠 Prompt Strategy  
Built with structured prompt chains and Chain-of-Thought reasoning to enhance consistency and interpret nuance:

- **Clip logic**: Detects emotional tone and topic transitions to suggest high-retention segments  
- **Fallback behavior**: Requests missing metadata instead of hallucinating when episode inputs are incomplete  
- **Style variation**: Applies formatting and tone parameters for different social media personas  
- **Title optimization**: Uses emotional resonance scoring + packaging prompt chains

---

## 🛠️ Tech Stack  
- **GPT-4** (ChatGPT) – prompt-driven generation engine  
- **NoteGPT** + **YouTube Transcript API** – source ingestion  
- **n8n** – automation and routing  
- **JSON** – structured delivery format  
- **Google Sheets + Notion** – content pipeline and team handoff  
- **Frame.io** – video editing feedback loop  
- _(Claude used for prompt testing and alt caption generation)_

---

## 🚧 Status  
- ✅ Live in weekly production  
- 🟡 Refining prompt logic and JSON formatting  
- 🔴 Automation layer + public repo polish (in progress)

---

## 🛣️ What’s Next  
- Smarter clip detection based on episode themes and emotional arcs  
- End-to-end automation: JSON → publishing tool integration  
- AI-generated thumbnails and auto-packaged delivery folders

---

## 🤝 Contributions  
This is a solo case-study project designed for AI workflow architecture.  
Suggestions and feedback welcome.