BlogPostAI is an autonomous research-to-publication blog generation system.
Given a single user prompt, it independently researches the topic, extracts key information, and writes a long-form blog post with a controllable narrative style — without human back-and-forth.

This is not a chat assistant.
It is a pipeline.

---

What This Project Does

From one prompt, BlogPostAI:

1. Plans research
2. Searches the web
3. Fetches and cleans sources
4. Summarizes relevant information
5. Extracts writing style
6. Writes a complete blog post
7. Renders the output in a UI

All steps are automated.

---

Core Architecture

User Prompt ↓ Research Planner (LLM) ↓ Search Engine ↓ Fetch HTML ↓ Clean Text ↓ Summarizer (LLM) ↓ Style Controller (LLM) ↓ Writer (LLM) ↓ Rendered Blog Output

Each stage is isolated and replaceable.

---

Project Structure

app/ ├── llms/ │ ├── research_planner.py # Decides what to research │ ├── summarizer.py # Extracts key points from sources │ ├── style_controller.py # Infers tone, narrative style, audience │ └── writer.py # Writes the final blog post │ ├── research/ │ ├── search.py # Runs search queries │ ├── fetch.py # Fetches raw HTML │ └── clean.py # Cleans text for LLMs │ ├── pipeline/ │ └── run.py # Orchestrates the full pipeline │ ├── utils/ │ ├── render_blog.py # Converts text → displayable output │ └── validate_images.py # (Optional) image validation │ └── ui/ └── streamlit.py # Streamlit UI

---

Key Design Decisions

1. Pipeline Over Chat

This system is deterministic and one-shot.
No conversational memory. No backtracking.

---

2. LLMs Are Used Only Where They Add Value

LLMs are used for:

• Research planning
• Summarization
• Style inference
• Writing

Everything else is traditional code.

---

3. Style Is Decoupled From Writing

The writer does not decide tone or structure.
Style is inferred once and passed in.

This prevents prompt drift and contradictions.

---

4. Strict JSON Output Was Intentionally Removed

Early versions enforced JSON outputs.

This was removed because:

• It caused frequent failures
• It added no real value for long-form writing
• Natural language output is more reliable for blogs

---

What This Is Not

• ❌ A chatbot
• ❌ A SEO keyword stuffer
• ❌ A web scraper farm
• ❌ A prompt-engineering toy
• ❌ A Medium clone

---

Current Capabilities

• Reverse-chronological storytelling
• First-person or third-person narration
• Emotional or factual tone
• Research-backed or experience-driven writing
• Fully autonomous execution
• UI-based execution via Streamlit

---

Known Limitations (Intentional)

• Scraping may fail on paywalled or protected sites (403s)
• Research depth depends on source availability
• No plagiarism detection
• No fact verification beyond retrieved sources

---

How to Run

Install dependencies

pip install -r requirements.txt

Run the Streamlit UI

streamlit run app/ui/streamlit.py


⸻

Example Prompt

Write me a blog about my trip to Disney.
Start from when I left and go backwards.
My favorite ride was the Incredicoaster,
while my kids loved Guardians of the Galaxy.
I want a lot of emotions and vivid imagery.

Output: a complete, styled blog post — no follow-ups required.

⸻

Project Status

MVP complete.

Further work would be:
	•	UX polish
	•	Export formats
	•	Caching
	•	Style presets

None are required for core functionality.

⸻

Why This Exists

This project explores autonomous content generation as a system — not as a chat trick.

Writing is treated as a pipeline problem, not a conversation.

⸻

License

MIT

---

If you want:
- a **short recruiter README**
- a **startup-style README**
- or a **1-page stripped version**

say the word.