How ssdeb-rr/linkedin-post-generator works
Project type and purpose
• This is a full-stack AI tool (Python-based) with a lightweight Streamlit UI for generating LinkedIn posts.
• Purpose: Ingest raw post data, extract metadata (topic, tone, length), and generate new influencer-style posts via a fine-tuned LLM (LLama 3.3 on Groq).
File structure
Top-level elements:
• data/
└─ raw_posts.json, processed_posts.json (input/output datasets)
• few_shot.py (builds few-shot prompts)
• llm_helper.py (abstracts calls to Groq Console / Langchain)
• preprocess.py (parses raw_posts.json → metadata)
• post_generator.py (assembles prompt + calls LLM to generate new post)
• main.py (Streamlit front-end glue)
• requirements.txt, .gitignore (env & dependencies)
README for additional insights
• Two main stages at runtime:
Stage 1: Metadata extraction (preprocess.py)
Stage 2: AI post generation (post_generator.py + few_shot.py + llm_helper.py)
• UI: Streamlit (main.py)
• External dependency: Groq Console API, Langchain orchestration, LLama 3.3 model.
Detailed coverage
• Don’t omit the Chain-of-Thought prompt logic in few_shot.py—show it as a sub-block of the prompt builder
• Extension points: switching to a database, adding a caching layer, integrating new LLMs